US20110125508A1

Movatterモバイル変換

Info

Publication number: US20110125508A1
Application number: US12/994,716
Authority: US
Inventors: Peter Kelly; Michael Raymond Reynolds; Christopher John Joseph Sutton
Original assignee: Intrasonics SARL
Current assignee: Intrasonics SARL
Priority date: 2008-05-29
Filing date: 2009-05-29
Publication date: 2011-05-26
Also published as: WO2009144470A1; CN102047324A; MX2010013076A; BRPI0913228A2; BRPI0913228B1; EP2325839A1; US8560913B2; GB2460306A; DK2631904T3; JP2011523091A; GB2460306B; US20120004920A1; CN102881290A; ES2545058T3; EP2631904B1; GB0821841D0; BRPI0913228A8; CN102881290B; EP2301018A1; PL2631904T3

Abstract

A data hiding system is described for hiding data within an audio signal. The system can be used for watermarking, data communications, audience surveying etc. The system hides data in an audio signal by adding artificial echoes whose polarity varies with the data to be hidden. In one embodiment, each data value is represented by a positive and a negative echo having different delays. A receiver can then remove the effects of natural echoes and/or periodicities in the audio signal by differencing measurements obtained at the different delays.

Description

The present invention relates to a system for embedding data in an audio signal and to its subsequent recovery, which can be used for watermarking, data communications, audience surveying etc. The invention has particular relevance, to a system for hiding data in an audio signal by adding echoes to the audio signal and to a system for recovering the hidden data by detecting the added echoes.
U.S. Pat. No. 5,893,067 discloses a technique for hiding data within an audio signal for transmission to a remote receiver. The data is hidden in the audio signal by adding an artificial echo to the audio signal and varying the amplitude and/or delay of the echo in accordance with the data to be hidden.
A problem with the data hiding technique described in U.S. Pat. No. 5,893,067 is that in most situations, natural echoes can mask the artificial echoes making it difficult for the receiver to be able to identify the artificial echoes and hence recover the hidden data.
The present invention aims to provide an alternative data hiding technique. One embodiment of the invention at least alleviates the above problem by applying echoes of opposite polarity to represent each data value.
According to one aspect, the present invention provides a method of embedding a data value in an audio signal, the method comprising: generating an echo of at least a portion of the received audio signal; and embedding the data value in the audio signal by combining the received audio signal with the generated echo; wherein the data value is embedded in the audio by varying the polarity of the echo that is combined with the audio signal in dependence upon the data value. The inventors have found that using polarity modulation to embed the data in the audio signal can make the recovery of the embedded data easier in the receiver, especially in the presence of natural echoes caused, by for example the acoustics of the room. The polarity modulation can be achieved by varying the echo that is generated and/or by varying the way in which the echo is combined with the audio signal.
In one embodiment, the generating step generates a first echo of at least a portion of the received audio signal and a second echo of at least a portion of the received audio signal, the first and second echoes having first and second polarities respectively, which polarities vary in dependence upon the data value; and wherein the embedding step embeds the data value in the audio signal by combining the received audio signal with the generated first and second echoes.
Each of the echoes may be generated by repeating at least a part of said audio signal. The first echo may be generated by repeating a first portion of the audio signal and the second echo may be generated by repeating a second portion of said audio signal. Alternatively, the first and second echoes may be generated by repeating substantially the same first portion of the audio signal. Where the audio signal is received as a stream of samples, the or each echo may be generated by passing the stream of audio samples through a delay line.
In one embodiment, third and fourth echoes may be generated, the third echo having the same polarity as said second echo and the fourth echo having the same polarity as said first echo. In this case, the third and fourth echoes may be generated by repeating substantially the same second portion of the audio signal which is different to the first portion repeated by the first and second echoes. The second portion of the audio signal may be adjacent to the first portion. The generating step may generate the third and fourth echoes by inverting the polarity of a gain factor applied to the echoes before being combined with the audio signal.
The first echo may be combined with the audio signal at a first delay relative to the first portion of the audio signal; the second echo may be combined with the audio signal at a second delay relative to said first echo; the third echo may be combined with said audio signal at a third delay relative to said second portion of the audio signal; and the fourth echo may be combined with the audio signal at a fourth delay relative to the third echo. The first delay may be equal to said third delay and/or the second delay may be equal to said fourth delay. In one embodiment, the delays and the amplitudes of the echoes are independent of the data value.
Preferably the first and third delays are between 0.5 ms and 100 ms and more preferably between 5 ms and 40 ms; and the second and fourth delays delayed (relative to the first and third echoes respectively) by between 0.125 ms and 3 ms and more preferably between 0.25 ms and 1 ms, as these delays are similar to those of natural echoes and so are less noticeable to users. In one embodiment, the or each echo has an amplitude that is less than the amplitude of said audio signal. Preferably the or each echo is faded in and out to reduce obtrusiveness of the echoes to a listener.
The first and second portions of the audio signal should be long enough for the receiver to be able to detect the presence of the echoes but not too long as to overly reduce the data rate that can be communicated. The inventors have found that echoes having durations of between 20 ms and 500 ms provides a reasonable data rate whilst keeping to a minimum data transmission errors when transmission occurs over an acoustic link. If transmission is over an electrical link, then shorter echoes may be used.
The echoes may be combined with the audio signal by adding and/or subtracting the echoes to/from the audio signal. The polarity of each echo can therefore be controlled by controlling the way in which each echo is combined with the audio signal.
This aspect of the invention also provides a computer implementable instructions product comprising computer implementable instructions for causing a programmable computer device to carry out the method described above.
This aspect also provides an apparatus for embedding a data value in an audio signal, the apparatus comprising: an echo generator operable to generate an echo of at least a portion of the received audio signal; and a combiner operable to combine the received audio signal with the generated first and second echoes to embed the data value in the audio signal; wherein the echo generator and/or the combiner are arranged so that the data value is embedded in the audio by varying the polarity of the echo that is combined with the audio signal in dependence upon the data value.
According to another aspect, the present invention provides a method of recovering a data value embedded in an audio signal, the method comprising: receiving an input signal having the audio signal and an echo of at least part of the audio signal whose polarity depends upon said data value; processing the received input signal to determine the polarity of the echo; and recovering the data value from the determined polarity.
The input signal may comprise a first echo of at least a portion of the audio signal and a second echo of at least a portion of the audio signal, the first and second echoes having first and second polarities respectively, which polarities vary in dependence upon the data value; and wherein the processing step processes the input signal to combine the first and second echoes and to determine the polarity of the combined echoes and wherein the recovering step recovers the data value from the determined polarity of the combined echoes.
In one embodiment the processing step processes the input signal to determine a first autocorrelation measure that depends upon the first echo and a second autocorrelation measure that depends upon the second echo and combines the echoes by differencing the first and second autocorrelation measures and determines the polarity of the combined echoes by determining the polarity of the result of the differencing step.
The first echo may be of a first portion of the audio signal and the second echo may be of a second portion of the audio signal. Alternatively the first and second echoes may be repeats of substantially the same portion of the audio signal.
In one embodiment, the input signal comprises first, second, third and fourth echoes, the first and fourth echoes having the same polarity and the second and third echoes having the same polarity which is opposite to the polarity of the first and fourth echoes, wherein the processing step processes the input signal to combine the first to fourth echoes and to determine the polarity of the combined echoes and wherein the recovering step recovers the data value from the determined polarity of the combined echoes.
In this embodiment, the processing step may process the input signal to determine a first autocorrelation measure that depends upon the first echo, a second autocorrelation measure that depends upon the second echo, a third autocorrelation measure that depends upon the third echo and a fourth autocorrelation measure that depends upon the fourth echo and combines the echoes by differencing the autocorrelation measures and determines the polarity of the combined echoes by determining the polarity of a result of the differencing step.
The differencing step may perform a first difference of the first and third autocorrelation measures, a second difference of the second and fourth autocorrelation measures, a third difference of the result of said first difference and the result of the second difference and wherein the polarity of the combined echoes may be determined from the polarity of a result of the third difference.
The first and second echoes may be repeats of substantially the same first portion of the audio signal and the third and fourth echoes may be repeats of substantially the same second portion of the audio signal. Alternatively, the first and third echoes may be repeats of substantially the same first portion of the audio signal and the second and fourth echoes may be repeats of substantially the same second portion of the audio signal. In one embodiment, the or each echo is faded in and out to reduce obtrusiveness of the echoes to a listener. In this case, the polarity of the echo may be determined when the amplitude of the echo is at or near a maximum.
The first echo may be delayed relative to said first portion of the audio signal by a first delay; the second echo may be delayed relative to the first echo by a second delay; the third echo may be delayed relative to the second portion of the audio signal by a third delay; and the fourth echo may be delayed relative to the third echo by a fourth delay. The first delay may be equal to the third delay and/or the second delay may be equal to said fourth delay.
According to this aspect, a computer implementable instructions product is also provided comprising computer implementable instructions for causing a programmable computer device to carry out the above method.
This aspect also provides an apparatus for recovering a data value embedded in an audio signal, the apparatus comprising: an input for receiving an input signal having the audio signal and an echo of at least part of the audio signal whose polarity depends upon said data value; a processor operable to process the input signal to determine the polarity of the echo; and a data regenerator operable to recover the data value from the determined polarity.

These and other aspects of the invention will become apparent to those skilled in the art from the following detailed description of exemplary embodiments, which are described with reference to the following drawings in which:

FIG. 1 is a block diagram illustrating the main components of a transmitter and receiver used in an exemplary embodiment;

FIG. 2ais an impulse plot illustrating the echoes that are added to an audio signal to encode a binary “one”;

FIG. 2bis an impulse plot illustrating the echoes that are added to an audio signal to encode a binary “zero”;

FIG. 3ais an impulse plot illustrating the presence of artificial echoes for a binary “one” after Manchester encoding and illustrating natural echoes;

FIG. 3bis an impulse plot illustrating the presence of artificial echoes for a binary “zero” after Manchester encoding and illustrating natural echoes;

FIG. 4 is a block diagram illustrating in more detail the encoding performed in the transmitter shown inFIG. 1;

FIG. 5 is a block diagram illustrating the main components of an echo generation and shaping module forming part of the transmitter shown inFIG. 1;

FIG. 6aillustrates a shaping and modulation function that is applied to the echoes prior to being combined with the audio signal when a binary “one” is to be transmitted;

FIG. 6billustrates a shaping and modulation function that is applied to the echoes prior to being combined with the audio signal when a binary “zero” is to be transmitted;

FIG. 6cillustrates the way in which the shaping and modulation function varies when two successive binary “ones” are to be transmitted;

FIG. 6dillustrates the shaping and modulation function that is applied when a binary “zero” is transmitted after a binary “one”;

FIG. 7 illustrates the processing performed in the receiver shown inFIG. 1 for recovering the hidden data from the received audio signal;

FIG. 8ais an autocorrelation plot for a typical audio signal without artificial echoes;

FIG. 8bis an autocorrelation plot for the audio signal with artificial echoes during a first half of a bit symbol;

FIG. 8cis an autocorrelation plot for the audio signal with artificial echoes during the second half of the bit symbol;

FIG. 8dis a plot obtained by subtracting the autocorrelation plot shown inFIG. 8cfrom the autocorrelation plot shown inFIG. 8b;

FIG. 9 is a block diagram illustrating an alternative form of receiver used to receive and recover the hidden data embedded in the audio signal;

FIG. 10 is a plot illustrating the way in which an FEC error count varies during a synchronisation process used to find the hidden data message within the input signal; and

FIGS. 11aand11billustrate the processing performed respectively by an FEC encoder and an FEC decoder in one embodiment.

OVERVIEW

FIG. 1 is a block diagram illustrating a transmitter and receiver system according to one embodiment in which atransmitter1 transmits data hidden within anacoustic signal3 to aremote receiver5. Thetransmitter1 may form part of a television or radio distribution network and the receiver may be a portable device such as a cellular telephone handset that is capable of detecting theacoustic signal3 output by thetransmitter1.

The Transmitter

As shown inFIG. 1, thetransmitter1 includes a forward error and correction (FEC)encoder module7, which receives and encodes the input data to be transmitted to theremote receiver5. The encoded message data output from theFEC encoding module7 is then passed to an echo generation and shapingmodule9, which also receives an audio signal in which the encoded message data is to be hidden. The echo generation and shapingmodule9 then hides the message data into the audio by generating echoes of the audio which depend upon the message data to be transmitted. The generated echoes are then combined with the original audio signal in acombiner module11 and the resulting modified audio signal is then passed to again control module13 for appropriate gain control. The audio signal is then converted from a digital signal to an analogue signal by the digital toanalogue converter15 and it is then amplified by adriver module17 for driving aloudspeaker19 which generates theacoustic signal3 having the data hidden therein.

As will be described in more detail below, in this embodiment, the polarity of the echoes (as opposed to their lag and/or amplitude) is varied in order to encode the data to be transmitted. The inventors have found that this polarity modulation can be more robust in the presence of natural echoes and periodicities in the audio signal. This is particularly the case when each data value is represented by two echoes of the same magnitude but having different lags and opposite polarities. The polarities of the echoes representing each message bit are reversed to distinguish between a binary zero and a binary one. This is illustrated by the impulse plots illustrated inFIG. 2. In particular,FIG. 2ais an impulse plot illustrating the component signals that are present when a binary one is to be transmitted andFIG. 2bis an impulse plot illustrating the component signals present when a binary zero is to be transmitted. As shown inFIG. 2a, the component signals include aninitial impulse21 representing the original audio signal followed by two lower amplitude impulses23-1 and23-2 representing the two echoes of theoriginal signal component21 which are added to the audio signal. As can be seen by comparingFIGS. 2aand2b, when a binary one is to be transmitted, a positive echo23-1 is transmitted first followed by a negative echo23-2; and when transmitting a binary zero a negative echo23-1 is transmitted first followed by a positive echo23-2. Although this could be reversed if desired.

As shown inFIG. 2, in this embodiment, the first echo is added with a lag of approximately ten milliseconds and the second echo is added 0.25 milliseconds after the first echo. This is the same regardless of whether a binary one or a binary zero is to be transmitted. Additionally, as represented inFIG. 2, in this embodiment, the echoes that are added have lower amplitudes compared with the amplitude of the original audio signal. In particular, in this embodiment, the amplitude of the echoes is approximately one third that of the original audio signal.

The Receiver

FIG. 1 also illustrates the main components of thereceiver5. As shown, the receiver includes a microphone31 for detecting theacoustic signal3 and for converting it into a corresponding electrical signal which is then filtered and amplified by filter andamplification circuitry33. The output from thefilter amplification circuitry33 is then digitised by an analogue todigital converter35 and the digital samples are then passed to anecho detector37. Theecho detector37 then processes the digital samples to identify the polarities of the echoes in the received signal. This information is then passed through adata recovery module39 which processes the echo information to recover the encoded message data. This message data is then decoded by adecoder41 to recover the original data that was input to the FEC decoding module of thetransmitter1.

Manchester Encoding

As will be explained in more detail below, theecho detector37 detects the echoes from the received signal by calculating the auto-correlation of the received signal at specified delays. However, natural echoes (e.g. room echoes) will also contribute to the autocorrelation values thus calculated as will periodicities of the original audio track. In order to distinguish the artificial echoes representing the encoded data from these natural echoes, the message data is also Manchester encoded so that a message data value of “1” is transmitted as a “1”, followed by a “0” (or vice versa), whilst a message data value of “0” is transmitted as a “0” followed by a “1”. In this embodiment, this Manchester encoding is performed by the echo generation and shapingmodule9. Therefore, when a message bit value of “0” is to be transmitted, for the first half of the symbol, the first echo23-1 is of positive polarity and the second echo23-2 is of negative polarity, whilst for the second half of the symbol, the first echo23-1 is of negative polarity and the second echo23-2 is of positive polarity. To transmit a message bit value of “0”, all polarities are reversed, as summarised in the table given below.


		first half of symbol	second half of symbol

data	polarity of	polarity of	polarity of	polarity of
value	first echo	second echo	first echo	second echo

0	Positive	negative	negative	positive
1	Negative	positive	positive	negative

The reason that the Manchester encoding can help to distinguish the artificial echoes from the natural echoes is that the natural echoes will be stable over the two half symbol periods. Therefore, by subtracting the autocorrelations in the second half of the symbol from autocorrelations in the first half of the symbol (or vice versa), the effect of the natural echoes and periodicities will cancel, whilst the autocorrelation peaks caused by the artificial echoes will add constructively. Similarly, the reason for using two echoes in each half symbol period is to distinguish the artificial echoes from periodicities in the original track. Typically, the autocorrelation of the original track will not change significantly between these two lags (i.e. between 10 ms and 10.25 ms). Therefore, by differencing the autocorrelations at the two lags, the effect of the periodicities is reduced and the autocorrelation peaks caused by the two echoes add constructively.

FIGS. 3aand3bare impulse plots showing the two half symbols and the artificial echoes23 that are added within each half symbol period to represent a binary “1” and a binary “0” respectively.FIGS. 3aand3balso illustrate natural echoes25-1 and25-2 which do not change from one half period to the next. Therefore, by subtracting the echoes in one half of the symbol period from the corresponding echoes (i.e. those with the same lag or delay) in the other half of the symbol period, the effect of the natural echoes and periodicities will cancel, whilst the artificial echoes will add constructively, thereby making it easier to detect the hidden data.

The above description provides an overview of the encoding and decoding techniques used in the present embodiment. A more detailed description will now be given of the main components of thetransmitter1 andreceiver5 to carry out the encoding and decoding processes described above.

FEC Encoder

FIG. 4 is a block diagram illustrating the main components of theFEC encoder module7 used in this embodiment. As shown, the first encoding module is a Reed-Solomon encoder module51 which uses a shortened (13,6) block code to represent the input data. The data output from the Reed-Solomon encoder51 is then passed to aconvolutional encoder53 which performs convolutional encoding on the data. The data bits output from theconvolutional encoder53 are then interleaved with each other by adata interleaving module55 to protect against errors occurring in bursts. Finally, a synchronisationdata adder module57 adds a sequence of synchronisation bits that will help thereceiver5 lock on to the encoded data within the receivedacoustic signal3. The output from the synchronisationdata adder module57 represents the message data which is then passed to the echo generation and shapingmodule9 shown inFIG. 1.

Echo Generation and Shaping

FIG. 5 is a block diagram illustrating the main components of the echo generation and shapingmodule9 and thecombiner module11 shown inFIG. 1. The input audio signal is represented by the sequence of audio samples a(n) which are applied to a 10millisecond delay unit61 and to the adder63 (corresponding to thecombiner11 shown inFIG. 1). The 10millisecond delay unit61 delays the input sample a(n) by 10 milliseconds which it then outputs to a 0.25millisecond delay unit65 and to asubtractor67. The 0.25millisecond delay unit65 delays the audio sample output from the 10millisecond delay unit61 by a further 0.25 milliseconds which it then outputs to thesubtractor67. Thesubtractor67 subtracts the 10.25 millisecond delayed sample from the 10 millisecond delayed sample outputting the result to amultiplier69. The delay units and the subtractor operate each time a new audio sample a(n) arrives. In this embodiment, the audio sample frequency is one of 8 kHz, 32 kHz, 44.1 kHz or 48 kHz.

Therefore, as those skilled in the art will appreciate, the 10millisecond delay unit61, the 0.25millisecond delay unit65 and thesubtractor67 will generate the two echoes23-1 and23-2 illustrated inFIG. 2. At this stage, however, the echoes that have been generated do not depend on the data to be transmitted. As will be explained below, this dependency is achieved by multiplying the echoes in themultiplier69 with a modulation function g(n) that is output by a lookup table71 which is addressed by lookuptable address logic73 in response to the current message data value. In particular, the lookup table output g(n) changes the polarity of the echoes in dependence upon the message data so that the echoes with the modulated polarities can then be added back to the original audio signal by theadder63 to generated the echo-modulated audio output signal.

Lookup Table Output g(n)

The inventors have found that abrupt changes in the echoes that are added can make the echoes more obtrusive to users in the vicinity of theloudspeaker19. Therefore, the lookup table output g(n) is gradually increased and decreased so that the echoes are effectively faded in and out.

Additionally, in this embodiment, the lookup table output g(n) also performs the above described Manchester encoding of the message data. The way in which this is achieved will now be explained with reference toFIG. 6. In particular,FIG. 6ais a plot illustrating the way in which the lookup table output g(n) varies over one symbol period, when the bit value of the message data is a binary “1”. In this embodiment, the symbol period is 100 ms. As shown, during the first half of the symbol period, the function g(n) increases from zero to a maximum value and then decreases back to zero at the end of the first half of the symbol period. During the second half of the symbol period, the function g(n) is negative and increases in magnitude to a maximum negative value and then decreases back to zero. As can be seen fromFIG. 6a, in this embodiment, the gradual increasing and decreasing of the lookup table output g(n) is achieved by using a sinusoidal function. Therefore, during the first half of the symbol, the combined echoes output from thesubtractor67 will be multiplied by a positive value and so their polarity will not be changed when they are multiplied by g(n) in themultiplier69. On the other hand, during the second half of the symbol period the lookup table output g(n) is negative and therefore, the polarities of the echoes output from thesubtractor67 will be reversed when the echoes are multiplied by g(n) in themultiplier69.

As mentioned above, the artificial echoes23 that are generated and added to the audio signal have an amplitude which is approximately a third that of the audio signal. In this embodiment, the amplitude of the echoes is controlled by the output of the lookup table g(n). As shown inFIG. 6a, the peak amplitude of the lookup table output g(n) is a third, which means that the maximum amplitude of the echoes which are added to the audio signal will be a third of the amplitude of the original audio signal.

As shown inFIG. 6b, when the message data is a binary value “0” the lookup table output g(n) is inverted compared with when the message data has a binary value of “1”. Therefore, during the first half symbol period, the polarity of the echoes output from thesubtractor67 will be reversed when they are multiplied by g(n) in themultiplier69 and during the second half of the symbol period the polarities of the echoes output by thesubtractor67 will not be inverted when they are multiplied by g(n) in themultiplier69.

FIG. 6cillustrates the lookup table output g(n) over two symbol periods when the message data to be transmitted is a binary “1” followed by another binary “1”. As shown inFIG. 6c, in this case, the lookup table output g(n) is a simple repeat of the output illustrated inFIG. 6a. Similarly, if successive values of the message data are binary “0's” then the lookup table output g(n) over the two symbol periods will be the inverse of that shown inFIG. 6c.

However, If the message data transitions from a binary “1” to a binary “0”, then instead of using a lookup table output function obtained by concatenating the functions shown inFIG. 6aandFIG. 6b, the function shown inFIG. 6dis used instead. As can be seen inFIG. 6d, when the lookup table output g(n) reaches its peak negative value in the first symbol period, it remains at that value until the peak would have occurred in the second symbol period before decreasing in magnitude back to zero. Similarly, when successive bits of the message data transition from a binary “0” to a binary “1”, the lookup table output g(n) over the two symbol periods will be the inverse of that shown inFIG. 6d. The inventors have found that not returning to the zero level in this way reduces the obtrusiveness of the echo modulation scheme that is used. This is because the human ear is more sensitive to changing echoes than to constant echoes.

As those skilled in the art will appreciate, the lookuptable address logic73 is responsible for analysing the successive bits of the message data and then to look up the appropriate part of the lookup table71 so that the appropriate output function g(n) is applied to themultiplier69.

Echo Detector

FIG. 7 is a part schematic and part block diagram illustrating the processing performed by theecho detector37. In particular,FIG. 7 illustrates 100 milliseconds of aninput signal61 at the input of theecho detector37. As those skilled in the art will appreciate, theinput signal61 is illustrated schematically as a continuous signal for ease of understanding but it will be a sampled and digitised waveform.

As illustrated by window i and window j, theecho detector37 includes two sliding windows63-1 and63-2 which extract adjacent segments of the input audio signal61-1 and61-2, each of length 50 milliseconds. Therefore, the twowindows63 extract portions of the inputacoustic signal61 which correspond to the above-described half symbol periods. As shown inFIG. 7, the extracted portion61-1 of the input acoustic signal is input to a first autocorrelation unit65-1 and the extracted portion61-2 of the input audio signal is input to a second autocorrelation unit65-2. Bothautocorrelation units65 operate to determine the autocorrelation of the corresponding portion61-1 or61-2 of the input acoustic signal at 10 millisecond and 10.25 millisecond lags. The determined autocorrelation values at lags 10.25 from autocorrelation units65-1 and65-2 are then input to asubtractor67, that subtracts the autocorrelation value obtained from window j from the autocorrelation value obtained from window i (or vice versa). The result of this subtraction is then supplied to anothersubtractor69. Similarly, the autocorrelation value atlag 10 milliseconds from window i and the autocorrelation value atlag 10 milliseconds from window j are output from theautocorrelation units65 to thesubtractor71, that subtracts the autocorrelation value obtained from window j from the autocorrelation value obtained from window i (or vice versa) and feeds the result to thesubtractor69. Thesubtractor69 then subtracts the output fromsubtractor67 from the output from subtractor71 (or vice versa). Therefore, the output from thesubtractor69 is represented by the following equation:

(Aⁱ(10)−A^j(10))−(Aⁱ(10.25)−A^j(10.25))

As mentioned above, subtracting the autocorrelation values of one half symbol period from the corresponding autocorrelation values of the other half symbol period can reduce the effect of natural echoes in the inputacoustic signal61. This is because natural echoes are unlikely to change from one half symbol period to the next and so their effect will be constant in the autocorrelations that are calculated. Consequently, performing this subtraction will remove this common effect. Likewise, subtracting the autocorrelation values obtained from each half symbol period will reduce the effect of periodicities in the original audio signal. This is because in the 0.25 ms delay between the first echo and the second echo in the half symbol period, the effect of the periodicities on the autocorrelations will be approximately constant and so this subtraction will remove this common effect. This will now be described in more detail with reference toFIG. 8.

FIG. 8ashows anautocorrelation plot81 obtained from a typical audio signal without any artificial echoes. As shown, theautocorrelation plot81 has a peak at zero lag. However, because of periodicities in the audio signal and because of natural echoes, theautocorrelation plot81 does not tail off towards zero until about 15 milliseconds after the initial peak and exhibits local peaks and troughs in between.Peak82 illustrates such a local peak that may occur as a result of a natural echo being added to the audio signal.

FIG. 8billustrates anautocorrelation plot83 for the same audio signal after a positive echo has been added at a lag of 10 milliseconds and a negative echo has been added at a lag of 12 milliseconds (rather than at 10.25 ms so that the two echoes can be seen more clearly). As shown inFIG. 8b, as a result of the artificial echoes, theautocorrelation plot83 includes a peak85 at 10 milliseconds and a peak87 at 12 milliseconds. However, thepeak85 is masked somewhat by theearlier peak82 caused by a natural echo.

FIG. 8cillustrates theautocorrelation plot89 for the audio signal after the echoes have been added in the second half of the symbol period. As shown, theautocorrelation plot89 includes anegative peak91 at 10 milliseconds and apositive peak93 at 12 milliseconds.

Finally,FIG. 8dillustrates the autocorrelation plot that is obtained by subtracting the autocorrelation plot shown inFIG. 8cfrom the autocorrelation plot shown inFIG. 8b. As can be seen, the common peaks in the autocorrelation plots shown inFIGS. 8band8chave been removed, whilst the

complementary peaks

85 and91; and87 and93 have added together to create the combined

peaks

95 and97 respectively. As those skilled in the art will appreciate, it is therefore much easier to detect the

peaks

95 and97 because their values are much greater than the autocorrelation values at other lags. This effect is further enhanced by subtracting the autocorrelation value at 12 milliseconds from the autocorrelation value at 10 milliseconds. This will effectively add the two

peaks

95 and97 together to provide an even larger peak, which can then be detected by suitable thresholding. The value of the corresponding data value can then be determined from the polarity of the combined peak.

As those skilled in the art will appreciate, in this embodiment, theecho detector37 does not calculate the autocorrelation of the input signal over all lags. It only calculates the autocorrelation values at the lags where the artificial echoes have been added. The plots shown inFIG. 8 show the autocorrelation values over lags from 0 to 15 milliseconds. These plots therefore help to illustrate the effect of natural echoes and periodicities in the audio signal which can mask the artificial echoes that are added to encode the data.

Synchronisation

In this embodiment, thereceiver5 knows the duration of each half symbol period. This defines the length of the windows63-1 and63-2 used in theecho detector37. However, theecho detector37 initially will not be synchronised with the transmitted data. In other words, theecho detector37 does not know where each symbol period begins and ends or where the start of the message is located. Therefore, in this embodiment, theecho detector37 performs the above analysis as each new sample is received from the analogue to thedigital converter35. The output from thesubtractor69 is then analysed by thedata recovery module39 to determine the most likely symbol boundaries. The data recovery module then determines the location of the start of the message by finding the synchronisation bits that were added by thesynchronisation data adder57. At this point, thedata recovery unit39 can start to recover the whole message from the polarity of the autocorrelation values output from thesubtractor69.

Once synchronisation has been achieved, theecho detector37 will typically determine the autocorrelation measurements in the middle of each half symbol period, when the echo is expected to be at its peak amplitude and thedata recovery module39 will determine the bit value from the polarity of the output from thesubtractor69. Theecho detector37 may also take measurements just before and just after the middle of each half symbol period, to allow thedata recovery module39 to track the synchronisation.

The message data recovered by thedata recovery module39 is then input to theFEC decoding module41 where the message data is decoded (using the inverse processing of the FEC encoder7) to obtain the original input data that was input to theencoder7 of thetransmitter1.

Modifications and Alternatives

In the above embodiments, the data was hidden within an audio signal by employing a number of echoes whose polarity varied with the data value to be transmitted. These echoes were added to the original audio signal after appropriate delays. As those skilled in the art will appreciate, the echoes may be added before the original audio signal (pre-echoes), before and after the original audio signal or only after the original audio signal.

In the above embodiment, synchronisation bits were added to the data that was transmitted so that the decoder can identify the boundaries of each symbol period and the start and end of each message. The use of such synchronisation bits significantly increases the overall message length that has to be transmitted (in some cases by as much as 25%). Additionally, as the decoding of each bit is subject to noise, the matching is not perfect which can reduce the chances of a successful synchronisation. The inventors have realised, however, that the synchronisation bits are not required. In particular, the inventors have realised that theFEC decoding module41 will have higher error rates when theecho detector37 is not properly synchronised with the incoming data compared with its error rate when the echo detector is synchronised with the incoming data. Therefore, in the embodiment illustrated inFIG. 9, the error output generated by theFEC decoding module41 is used to control the synchronisation of the receiver to the incoming data.

More specifically, in this embodiment, theecho detector37 receives a block of samples corresponding to one or more symbol(s) and determines the optimum time within that block of samples to detect the echoes within the symbols. Multiple symbols may be required when Manchester encoding is used as a Manchester encoded “one” looks the same as a Manchester encoded “zero” with a time shift. Therefore, it may be necessary to consider a number of symbols to allow the symbol boundaries to be identified. The actual determination of the optimum time within the block of samples to detect the echoes may be determined by passing the block of samples through a matched filter (loaded with the expected signal pattern for one symbol period) and the time within the symbol when the absolute output (averaged over a number of successive symbols) is at a maximum is deemed to be the best time to sample the symbols. For example, if there are N samples per symbol, and the block of samples has M symbols, then the following values are calculated:

average(0)=1/M*(x(0)+x(N)+x(2N)+ . . . )

average(1)=1/M*(x(1)+x(N+1)+x(2N+1)+ . . . )

. . .

average(N−1)=1/M*(x(N−1)+x(2N−1)+x(3N−1)+ . . . )

where x(i) is absolute output of the matched filter for sample i. The largest average value thus determined identifies the best time to detect the echoeswithin the incoming signal during each symbol.

Theecho detector37 then uses the determined optimum time to detect echoes in that symbol and in the previous N−1 symbols of the input signal (where N is the number of symbols in the transmitted message). Thedata recovery module39 then determines, from the detected echoes, bit value(s) for each symbol and outputs the string of bits corresponding to the possible message to theFEC decoding module41. TheFEC decoding module41 then performs the inverse processing of theFEC encoder7 to regenerate a candidate input data codeword, which is stored in thebuffer93. TheFEC decoding module41 also outputs an error count indicating how many errors are identified in the candidate codeword, which it passes to acontroller91. In response, thecontroller91 compares the error count with a threshold value and if it is greater than the threshold, then thecontroller91 flushes the candidate codeword from thebuffer93. The above process is then repeated for the next received symbol in the input signal, until thecontroller91 determines that the error count is below the threshold. When it is, thecontroller91 instructs theFEC decoding module41 to accept the candidate codeword, which it then outputs for further use in thereceiver5. In effect, therefore, theecho detector37, thedata recovery module39 and theFEC decoding module41 all operate on a window of the input signal corresponding to the length of the transmitted message, which window is slid over the input signal until a point is found where the FEC error count is below a defined threshold—indicating the identification of the full message within the input signal.

FIG. 10 is a plot illustrating the way in which the FEC decoding module's error count99 is expected to change as thewindow101 is slid over aninput signal103 containing adata message105, with the minimum appearing at symbol S_N, when thewindow101 is aligned with thedata message105 in theinput signal103. The threshold (Th) level is then set to reduce the possibility that false minimums in the FEC error output count are considered as possible codewords, so that (in the ideal situation) only when thereceiver5 is properly synchronised (aligned) to the message data, will the FEC decoding module's error count reduce below the threshold in the manner illustrated inFIG. 10. Ideally, in this embodiment, the FEC encoding/decoding that is used is designed to keep the error rate of theFEC decoding module41 high except when thewindow101 is aligned with themessage data105 in theinput signal103. The inventors have found that this simple thresholding technique is sufficient to identify the location of the message data in theinput signal103. However, if more accurate detection is required, then further consideration can be made, varying the possible positions of the start and end of the message and looking for the positions that give the minimum FEC error count.

The above technique is useful for finding a single message in the input signal. Clearly, if a sequence of such data messages is transmitted, then the synchronisation timing determined for the first data message may be used to identify the synchronisation timing for the next data message.

One problem identified by the inventors with the synchronisation approach discussed above is that theFEC encoder7 often uses cyclic codewords (for example when using Reed Solomon block encoding) which means that a one bit shift in the codeword can also be a valid codeword. This is problematic because it can result in false detections of a codeword (a so called false positive) in theinput signal105. This problem can be overcome by reordering the bits of the codeword in theFEC encoder7 in some deterministic manner (for example in a pseudo random manner), and using the inverse reordering in theFEC decoder41. The processing that may be performed by theFEC encoder7 and by theFEC decoder41 in such an embodiment is illustrated inFIGS. 11aand11brespectively. As shown, theFEC encoder7 performs a cyclic encoding of the data (in this case Reed Solomon encoding111), followed by a pseudorandom reordering113 of the data. The reordered data is then convolutionally encoded115 and then interleaved117 as before. Similarly, theFEC decoding module41 initially de-interleaves121 the data and performsconvolutional decoding123. TheFEC decoding module41 then reverses123 the pseudo random data reordering performed by theFEC encoder7 and then performs the Reed Solomon decoding125. As those skilled in the art will appreciate, by performing this reordering of the data in this way, if there is a bit shift in the message data output by thedata recovery module39, then it is far less likely to result in a valid codeword and so the FEC error rate output is unlikely to trigger the false identification of a data message.

In the above embodiments, each data value was represented by four echoes—two echoes in each of two half symbol periods. As those skilled in the art will appreciate, each data value may be represented by any number of echoes in any number of sub-symbol periods. For example, instead of having two echoes within each half symbol period, each data value may be represented by a single echo in each half symbol period. In this case, the echoes in each half symbol period would preferably be of opposite polarity so that the same differencing technique can be used to reduce the effects of natural echoes. Indeed, the inventors have found that in some cases using two echoes of opposite polarity in each half symbol period can result in some frequency components within the original audio signal adding constructively with the echoes and some frequency components within the original audio signal adding destructively with the echoes. If a single artificial echo is added, then such distortions are less evident making the hidden data less noticeable to users in the acoustic sound that is heard.

As those skilled in the art will appreciate, representing each data value by one or more echoes in different sub-symbol periods, means that the echoes in each sub-symbol period will be a repetition of a different portion of the audio signal. If there is only one symbol period, then each data value will be represented by echoes of the same (or substantially the same) portion of the audio signal.

In the above embodiments, each data value was represented by a positive and a negative echo in a first half symbol period and by a positive and a negative echo in the second half symbol period. The positive and negative echoes in the first half symbol period allowed the receiver to reduce the effects of periodicities in the original audio signal which effect the autocorrelation measurements. The use of complementary echoes in adjacent half symbol periods allows the receiver to reduce the effect of natural echoes within the received audio signal, which might otherwise mask the artificial echoes added to represent the data. As those skilled in the art will appreciate, in other embodiments, neither or only one of these techniques may be used.

In the above embodiment, each data value was represented by echoes within two adjacent half symbol periods. As those skilled in the art will appreciate, these two half symbol periods do not have to be immediately adjacent to each other and a gap may be provided between the two periods if required.

In the above embodiment, the echoes in each half symbol period were of exactly the same portion of the audio signal. As those skilled in the art will appreciate, this is not essential. The echoes in each half symbol period may be of slightly different portions of the audio signal. For example, one echo may miss out some of the audio samples of the audio signal. Alternatively, the audio signal may include different channels (for example left and right channels for a stereo signal) and one echo may be formed from a repetition of the left channel and the other may be formed from a repetition of the right channel. With modern multi channel surround sound audio the repetitions can be of any of these channels.

In the above embodiment, the echoes generated within the transmitter were added to the original audio signal. As those skilled in the art will appreciate, the generated echoes may be combined with the original audio signal in other ways. For example, the echoes may be subtracted from the audio signal. Similarly, instead of inverting the echoes to be added to the audio (by controlling the polarity of the function g(n)), the same result can be achieved by changing the way in which the echoes are combined with the audio signal. For example, one echo may be added to the original audio signal whilst the next echo may be subtracted from the audio signal.

In the above embodiment, the lookup table stored values for g(n) corresponding to one or two bits of the message data (as illustrated inFIG. 6). As those skilled in the art will appreciate, this is not essential. For example, the lookup table could simply store a function which increased in value and then decreased in value. Additional circuitry could then be provided to convert the polarity of this output as appropriate for the two half symbol periods. In this way, the function stored in the lookup table would only control the fading in and out of the echo and the additional circuitry would control the polarity of the echo as required.

In the above embodiment, the Manchester encoding was performed by the echo generation and shaping module. As those skilled in the art will appreciate, this Manchester encoding, if performed, may be performed within the FEC encoding module.

As those skilled in the art will appreciate, the techniques described above for hiding data within the audio may be done in advance of the transmission of the acoustic signal or it may be done in real time. Even in the case where the data is to be embedded within an audio signal in real time, some of the processing can be done in advance. For example, the FEC encoding may be performed on the data in advance so that only the echo generation and echo shaping is performed in real time.

In the above embodiments, specific examples have been given of sample rates for the audio signal and symbol rates for the data that is hidden within the audio signal. As those skilled in the art will appreciate, these rates are not intended to be limiting and they may be varied as required. However, in order to keep the obtrusiveness of the added echoes to a minimum, the data rate of the encoded data is preferably kept between one and twenty symbols per second. This corresponds to a symbol period of between 50 ms and 1 second. In some embodiments, a long symbol period is beneficial because the added echoes will span across spoken words within the audio, making it easier to hide the data echoes within the audio. A larger symbol period also reduces audibility of the echoes. This is because humans are more sensitive to changing echoes than they are to static or fixed echoes. Therefore, by having a longer symbol period, the rate of change of the echoes is lower making the presence of the echoes less noticeable to a user.

In the above embodiment, the data rate of the data added to the audio signal in the transmitter was constant and was known by the receiver. This knowledge reduces the complexity of the receiver circuitry for locking on to the data within the received signal. However, it is not essential to the invention and more complex circuitry may be provided in the receiver to allow the receiver to try different data rates until the actual data rate is determined. Similarly, the receiver may use other techniques to synchronise itself with the transmitted data so that it knows where the symbol boundaries are in advance of receiving the data.

In the above embodiment, FEC encoding techniques were used to allow the receiver to be able to correct errors in the received data. As those skilled in the art will appreciate, such encoding techniques are not essential to the invention. However, they are preferred, as they help to correct errors that occur in the transmission process over the acoustic link.

In the above embodiments, the peak amplitudes of the echoes were all the same and were independent of the data value being transmitted. As those skilled in the art will appreciate, the peak amplitudes of the echoes may also be varied with data to be transmitted if desired.

In the above embodiment, the echoes in each half symbol period were at the same delays relative to the original audio signal. As those skilled in the art will appreciate, this is not essential. There may be some variation in the actual delay values used within each half symbol period.

In the above embodiment, the second echo within each half symbol period was generated by delaying the first echo by a further delay value. In an alternative embodiment, each echo within each sub-symbol period may be independently generated from the original audio signal using an appropriate delay line.

As those skilled in the art will appreciate, various uses can be made of the above communication system. For example, the encoded data may be used as a watermark to protect the original audio signal. Alternatively, the embedded data may be used to control the receiver so that it can respond in synchronism with the audio signal. In particular, the decoder can be programmed to perform some action a defined time after receiving the codeword. The time delay may be programmed into the decoder by any means and may even be defined by data in the received codewords. When used to perform such synchronisation, shorter symbol periods are preferred as shorter symbol periods allows for better temporal resolution and hence more accurate synchronisation. The data may be used for interactive gaming applications, audience surveying, e-commerce systems, toys and the like. The reader is referred to the Applicant's earlier International application WO02/45273 which describes a number of uses for this type of data hiding system.

In the above embodiment, the receiver performed autocorrelation measurements on the input audio signal in order to identify the locations of the echoes. As those skilled in the art will appreciate, other techniques can be used to identify the echoes. Some of these other techniques are described in the Applicant's earlier PCT application PCT/GB2008/001820 and in U.S. Pat. No. 5,893,067, the contents of which are incorporated herein by reference. Typically, although not necessarily, the techniques involve some form of autocorrelation of the original audio signal or of parameters obtained from the audio signal (eg LPC parameters, cepstrum parameters etc). As an alternative, a best fit approach could be used in which an expected audio signal (with different echo polarities) is fitted to the actual signal until a match is found and the polarity of the echoes thus determined.

In the embodiment described above, a single transmitter was provided together with a receiver. As those skilled in the art will appreciate, multiple transmitters and/or multiple receivers may be provided. Further, the components of the transmitter may be distributed among a number of different entities. For example, the encoding and data hiding part of the transmitter may be provided within a head end of a television distribution system or a user's set top box and theloudspeaker19 may be a speaker of the user's television set.

In the above embodiments, the echoes were directly derived from the original audio signal. In alternative embodiments, the echo may not include all frequency components of the audio signal. For example, one or more of the echoes may be generated from a portion of the audio signal after it has been filtered to remove certain frequencies. This may be beneficial where it is found, for example, that there is additional noise in the low frequency part of the echoes but not in the higher frequency part. In this case, the received signals would also be filtered to remove the lower frequency components (for example frequencies below about 500 Hz) so that only the higher frequency components (those above the lower frequency components) of the audio signal and the echoes would be present in the signals being analysed. Alternatively, in this case, the received signal may be passed through a filter that simply reduces the level of the lower frequency components in the received signal compared with the higher frequency components. This will have the effect of reducing the relevance of the noisy low frequency part of the received signal in the subsequent decoding process. Similarly, if it turns out that the added echoes introduce a noticeable distortion in the higher frequencies of the composite audio signal, then the echoes (or the signals from which they are derived) may be low pass filtered to remove the higher frequencies.

The division of the audio signal into separate frequency bands can also be used to carry data on multiple channels. For example, if the frequency band is divided into a high frequency part and a low frequency part, then one channel may be provided by adding echoes to the high frequency part and another channel may be provided by adding different echoes to the low frequency part. The use of multiple channels in this way allows frequency or temporal diversity if the data carried in the two channels is the same; or allows for an increased data transfer rate if each channel carries different data. Multiple channels can also be provided where the audio signal also contains multiple channels (used to drive multiple speakers). In this case, one or more data channels may be provided in the audio signal for each audio channel.

In the above embodiment, data was hidden within an audio signal by adding echoes to the audio signal. In some situations, the incoming audio may already contain hidden data in the form of such echoes. In this case, the encoder could decode the existing hidden data from the received audio signal and then use the decoded data to clean the audio signal to remove the artificial echoes defining this hidden data. The encoder could then add new echoes to the thus cleaned audio signal to hide the new data in the audio signal. In this way, the original hidden data will not interfere with the new hidden data.

In the above embodiment, the echoes were obtained by delaying digital samples of the audio signal. As those skilled in the art will appreciate, the echoes may be generated in the analogue domain, using suitable analogue delay lines and analogue circuits to perform the echo shaping and polarity modulation.

In the above embodiments, the audio signal with the embedded data was transmitted to a receiver over an acoustic link. In an alternative embodiment, the audio signal may be transmitted to the receiver over an electrical wire or wireless link. In such an embodiment, the data rates that are used may be higher, due to lower noise levels.

In the above embodiment, one data bit was transmitted within each symbol period. In an alternative embodiment, multiple bits may be transmitted within each symbol period. For example a second pair of echoes may be added at lags of 20 ms and 20.25 ms within each half symbol period to encode a second bit; a third pair of echoes may be added at lags of 30 ms and 30.25 ms within each half symbol period to encode a third bit etc. Each echo could then be faded in and out during each half symbol period and polarity modulated in accordance with the bit value as before. The fading in and out of the echoes for the different bits may be the same or it may be different for the different bits. The polarity modulation of the different echoes will of course depend on the different bit values to be transmitted in the symbol period. In a preferred embodiment, the echoes for the different bits within the same half symbol period are faded in and out at different times of the half symbol period, so that the different echoes reach their peak amplitudes at different times within the half symbol period. In this way, when the echo for one bit is at its peak amplitude (or when all the echoes for one bit are at their peak amplitudes—if there are multiple echoes representing each bit in each half symbol period), the echoes for the other bits will not be at their peaks. Doing this and sampling the different echoes when they are expected to be at their peak amplitudes, will reduce the interference between the echoes for the different bits within the same half symbol period. It also reduces constructive interference of the echoes that may render the added echoes more noticeable to a listener. Looking at this another way, this is the same as having multiple parallel data messages, each encoded as per the embodiments described above, but with their respective symbol periods offset in time from each other so that the echoes for the different messages peak at different times—thereby reducing interference between the messages if the echoes are all sampled around the time when they are each at their maximum amplitudes. This technique will increase the bit rate of data transmission between the transmitter and receiver. The additional bits may be of the same message or they may be bits of different messages.

The inventors have found that the above described data hiding techniques do not work as well during portions of the audio that include single tones or multiple harmonic tones, such as would be found in some sections of music. This is because the hidden data becomes more obtrusive to the listener in these circumstances and if the tones are being used as part of an automatic setup procedure they can cause the procedure to fail. Therefore, in one embodiment, the inventors propose to include (within the encoder) a detector that detects the level of tonality or other characteristic of the audio signal and, if it is highly tonal, that switches off the echo addition circuitry. Alternatively, as this switching off of the echoes may itself be noticeable to the user, the encoder may fade the echoes out during periods of high tonality and then fade them back in during periods of low tonality. In this way, the data is only added to the audio signal when the audio signal is not highly tonal in nature. Various techniques may be used for making this detection. One technique for determining the level of tonality of an audio signal (although for a different purpose) is described in the applicant's earlier PCT application WO02/45286, the contents of which are incorporated herein by reference. Another technique can be found in Davis P (1995) “A Tutorial on MPEG/Audio Compression”, IEEE Multimedia Magazine, 2(2), pp. 60-74. Instead of switching off the echo addition circuitry, the system may be arranged to adapt the amplitude of the added echoes depending on the detected characteristic of the audio signal. Alternatively, instead of varying the amplitudes of the echoes in this way, the encoder may instead or in addition vary the data rate or the symbol period in order to reduce the obtrusiveness of the hidden data during periods when the audio signal is highly tonal.

An embodiment was described above in which a single message was encoded and transmitted to a remote receiver as a number of echoes within an audio signal. In some applications, a sequence of messages may be transmitted. These messages may be the same or they may be different. In either case, each message may be transmitted after a preceding message has been transmitted. Alternatively, the end of one message may be overlapped with the start of the next message in a predefined way (so that the receiver can regenerate each message. This arrangement can increase the time diversity of the transmitted messages making them less susceptible to certain types of noise or data loss. In a further alternative, the data from the different messages may be interleaved in a known manner and transmitted as a single data stream to the receiver. The receiver would then regenerate each message by de-interleaving the bits in the data stream using knowledge of how the messages were originally interleaved.

As discussed above, Convolutional Coding is used as part of the forward error correction (FEC) encoder. As is well known to those skilled in the art, data encoded in this way generally is decoded using a Viterbi decoder, which operates by constructing a trellis of state probabilities and branch metrics. The transmitted data is often terminated with a number of zeros to force the encoder back to the zero state. This allows the decoder to start decoding from a known state, however, it requires extra symbols to be transmitted over the channel. An alternative technique is to ensure that the trellis start and end states are identical. This technique is referred to as tail biting and has the advantage of not requiring any extra symbols to be transmitted. Tail biting is used in many communications standards and, if desired, may be used in the embodiments described above.

The description above has described the operation of a system for hiding data as echoes within an audio signal. The systems described used time domain techniques to generate and add the echoes and to detect the echoes in the received signal. As those skilled in the art will appreciate, equivalent processing can be performed in the frequency domain to achieve the same or similar results.

The inventors have found that in some instances, the decoder does not work as well when the message consists of predominantly ‘zero’ bits (or conversely predominately ‘one’ bits), since under the encoding scheme an ‘all zeros’ codeword segment looks the same as a time-shifted ‘all ones’ codeword segment. A particular example is the ‘all zeros’ message, which results in an ‘all zeros’ codeword after Reed Solomon encoding. The encoding works best when there are approximately equal numbers of ones and zeros in the codeword, evenly distributed throughout the codeword. This can be achieved for the disclosed system by inverting the Reed Solomon parity bits. This has the effect of changing the all zeroes codeword to a mixture of zeroes and ones. This can also be achieved by altering the initial state of the feedback shift register used within the Reed Solomon encoder which is used to generate the parity bits. This gives more flexibility in setting the ratio of ones to zeroes in the codeword. Subsequent interleaving distributes these inverted parity bits throughout the codeword. As those skilled in the art of error detection and correction will appreciate, these approaches to balancing the distribution of ones and zeroes applies to any of the many FEC schemes implemented using feedback shift registers (or Galois field arithmetic) of which Reed Solomon is an example.

In the above embodiments, a number of processing modules and circuits have been described. As those skilled in the art will appreciate, these processing modules and circuits may be provided as hardware circuits or as software modules running within memory of a general purpose processor. In this case, the software may be provided on a storage medium such as a CD-ROM or it may be downloaded into an appropriate programmable device on a carrier signal over a computer network, such as the Internet. The software may be provided in compiled form, partially compiled form or in un-compiled form.