GB2436191A

GB2436191A - Predictive encoder/vector quantiser which transmits predictor state information for error checking and recovery

Info

Publication number: GB2436191A
Application number: GB0605050A
Authority: GB
Inventors: Jonathan Alastair Gibbs; Fikretler Halil
Original assignee: Motorola Inc
Current assignee: Motorola Solutions Inc
Priority date: 2006-03-14
Filing date: 2006-03-14
Publication date: 2007-09-19
Anticipated expiration: 2026-03-14
Also published as: GB0605050D0; WO2007106637A3; WO2007106637A2; GB2436191B

Abstract

The system includes predictive vector quantiser (VQ) comprising subtractor 110, quantiser 115, delaying dequantiser 125, adder 130, delay 140 and prediction matrix 145. The output 120 is the quantised difference between the input signal and the prediction, that is the prediction error. The receiver includes dequantiser comprising dequantiser 160, adder, delay 175 and prediction matrix 180 which mirror the encoder prediction operation and add the prediction error to recover the original signal. The problem with this system is that when an error occurs in channel 150 it affects all subsequent signals since it feeds into the prediction loop at the receiver and causes it to diverge from that in the transmitter. The invention subtracts the dequantised prediction error from the input signal 305 and feeds it through an inverse prediction matrix 310 to generate and approximation of the predictor state that is sent to the receiver. At the receiver the normal running prediction and the received prediction are used to decode the data and cross checked to detect errors. If one is found the running prediction state can be overwritten with the received predictor state to resynchronise it with the transmitter.

Description

--

COMMUNICATION UNIT, INTEGRATED CIRCUIT ND METHOD

THEREFOR

Field of the Invention

Embodiments of the present invention relate to source coding and methods for improving the performance of source coders and decoders in communication units operating over error prone channels. The invention is applicable to, but not limited to, improving the quality of speech, audio and video transmission systems.

Background of the Invention

Whereas traditionally, users of wireless digital communication systems have tolerated the effects of poor coverage on the conversational speech they have been listening to, this situation is changing. As they become more pervasive and begin to replace landline communications, wireless systems will increasingly be expected to convey speech, audio and video signals with little or no perceptually noticeable degradation due to channel errors. Channel errors, caused by signal distortion, are introduced into wireless transmitted signals primarily due to multi-path and fading effects.

Furthermore, cellular radio technology attempts to maximise use of the limited channel bandwidth by reusing the available frequencies in different cells. As a result, cellular radio systems also suffer from significant co-channel and adjacent channel interference in addition to the effects of multipath and fading.

The errors/distortion introduced into wireless transmitted signals result in received bits being wrongly decoded. Hence, the issue of data integrity, where error protection is required or desired, is therefore becoming increasingly important.

Many techniques may be used to reduce the impact of such errors/distortion; some of which expand the data bandwidth required, such as forward error correction coding (FEC) and others that do not expand the data bandwidth required, such as VQ index assignment.

Embedded source encoding involves a source being coded with one encoder, known as the core codec, and at the same time, additional data is transmitted alongside the core codec information to provide either increased fidelity or increased channel error resilience, or both.

The additional information may itself be structured in layers and be further embedded. Embedded coding is also sometimes referred to as scalable coding, often by those in the field of video source coding. The requirements for the core codec usually mean that efficient quantization schemes are required. However, then the problem becomes how to increase the resilience and/or quality of those schemes with additional data. Chen et al [2000] describe an embedded transform codec for speech, but without inter-frame prediction for error robustness, although this is not always possible or desirable.

Predictive source coding is well known in the art of source coding and compression, for example see A. Gersho and R.M. Gray "Vector Quantization and Signal Compression", Kluwer Academic Publishers, 1992, where adjacent observations of a source signal are relatively highly correlated. Such techniques are found in both video compression (motion compensation vectors) and speech compression (Linear Prediction and Long-Term prediction) Many of these schemes make use of autoregressive (AR) predictive quantization, which predicts the value to be quantized from previous estimates of that signal. AR prediction is well known to provide high coding gains with low complexity. However, AR predictive quantizers have infinite memory, and so have a tendency in channel error conditions for the decoder to diverge from the required value.

Referring now to FIG. 1, a known schematic of an auto regressive (AR) predictive quantizer is illustrated.

Predictive quantization is often used in source coding as a convenient way to exploit time correlation between adjacent samples, in order to achieve a coding gain. Two methods of predictive quantization are commonly employed; Auto-regressive (AR) and Moving Average (MA) prediction, with AR prediction often providing the most gain for a given predictor order. Unfortunately, AR predictive quantizers are highly sensitive to channel errors, since the errors may be propagated almost indefinitely. A known AR predictive quantizer is shown in FIG. 1.

Referring to the AR predictive quantizer in FIG. 1, presenLed with a given vector quantity 105, , to quantize, an error signal is derived by comparing it in subtract logic 110 to a predicted quantity from past quantized estimates of 105, a 135, using delayed quantizer logic 125, summing logic 130, a delay 140 and predictor matrix G' 145. Thus, the error signal is quantized in quantizer 115 and the index 1120 is transmitted across a transmission system.

If errors are introduced by the channel 150 during transmission, the received index will be incorrect and a value 1155 will be received. At the decoder, the reverse process is implemented. Here, delay quantizer logic 160 delays the value 1155, which is input to summing logic 165. The given vector quantity 105, , is derived by comparing it in summer logic 165 to a predicted quantity from past quantized estimates of 2 170, using delay 175 and predictor matrix G' 180 and looped into summing logic 165.

Various methods have been proposed to limit the error propagation in AR predictive quantizers due to the divergence introduced by bit errors by, for example, inserting occasional memoryless coded samples of the source signal into the bitstream. This method is commonly employed in video source encoders such as ITU-T Recommendations H.263 and H.264 to limit the error propagation problem or in the area of speech source encoding as described in J. Skogland and J. Linden, "Predictive VQ for Noisy Channel Spectrum Coding: AR or MA?", Proceedings of ICASSP 1997, pp1351-1354. However, although this ensures that the divergence is not infinite, the error events will almost always extend beyond the error event and the resumption of perfectly received data.

Thus, such solutions are not ideal and there exists a need to provide a communication unit comprising an encoder and/or decoder, an integrated circuit and a method of operation therefor.

Summary of the Invention

In accordance with aspects of the present invention, there is provided a source coding and compression unit, an integrated circuit and method of operation therefor, as defined in the appended Claims.

Brief Description of the Drawings

FIG. 1 illustrates a known schematic of an auto regressive (AR) predictive quantizer.

Exemplary embodiments of the present invention will now be described, by way of example only, with reference to the accompanying drawings, in which: FIG. 2 illustrates a source encoder-decoder arrangement adapted to support embodiments of the present invention; FIG. 3 shows a block diagram of a modified auto-regressive (AR) predictive quantizer arranged to support embodiments of the present invention; and FIG. 4 illustrates a flowchart of an auto-regressive (AR) predictive quantizer decoding operation according to embodiments of the present invention.

Description of Embodiments of the Present Invention In one embodiment of the present invention a communication unit is described comprising a predictive source encoder capable of representing an input signal.

The predictive source encoder comprises or is operably coupled to prediction memory, and quantization logic capable of representing the state of the prediction memory associated with the predictive source encoder.

The provision of quantization logic to represent the state of the prediction memory associated with the predictive source encoder provides the advantage of limiting the error propagation effects due to channel errors, which would otherwise cause the encoder and decoder predictors to become out of step with one another.

In one embodiment of the present invention the quantization logic may derive an independent signal by subtracting an effect of one or more quantization decision(s) of the predictive source encoder from the input signal. The independent signal may then be passed through a function performing an inverse of the predictor, which provides a signal that may be similar to the predictor state. Notably, and advantageously, this signal does not include quantization noise from past quantization decisions.

The provision of quantization logic to use an inverse of the predictor to derive a second signal to be quantized provides the advantage that when both quantization decision values are present at the decoder, from the core predictive vector quantizer and the state quantizer, two versions of the input signal with uncorrelated quantization noise may be derived. The first of these two versions of the input signal is the original predictive quantizer decoder output. The second version of the input signal is taken from a second predictive quantizer decoder provided with the same quantization decision(s) as the first decoder, having first been bootstrapped with the quantized value representative of the predictor memory of the encoder. When both of these decoders are provided with error-free data, their respective outputs may be weighted and summed to derive a version of the input signal with lower error variance, or quantization noise, than either of the previous two individual versions. This combination of these two versions of decoder output achieves a Multi-Description coding (MDC) gain.

In one embodiment of the present invention, the communication unit comprises combining logic to combine the predictive values of the input signal and the quantized memory state into a single embedded or scalable transmission. This has the advantage that in the event that a transmission channel has very low error rate, the data conveying the quantized predictor state may be omitted to save bandwidth, without significantly affecting the quality of the synthesised source.

In one embodiment of the present invention, a method of predictive source encoding in a communication unit is described. The method comprises encoding an input signal, in a predictive source encoder, to produce a first encoded signal; quantizing prediction memory associated with the predictive source encoder; encoding the quantized prediction memory to produce a second encoded signal; and transmitting a signal comprising both the first encoded signal and the second encoded signal.

The method may comprise subtracting an effect of one or more decision(s) of the predictive source encoder from the input signal and deriving an independent signal in response thereto.

In one embodiment, the method may further comprise using an inverse of a predictor of the predictive source encoder to derive the prediction memory to be quantized.

In an alternative embodiment, the method may comprise determining an optimum quantized value of prediction memory through an analysis by synthesis process. In this embodiment, the method may comprise minimizing an error, or perceptually weighted error, between a prediction from a quantized value of prediction memory and a derived independent signal.

In an alternative embodiment, the method may comprising determining whether computation logic in predictive source encoder is complex and in response to the determining: using an inverse of a predictor of the predictive source encoder to derive Lhe prediction memory to be quantized or determining an optimum quantized value of prediction memory through an analysis by synthesis process.

In one embodiment of the present invention a communication unit comprises two predictive source decoders capable of representing a received signal. A first predictive source decoder derives a predictor memory state from previous decoding decisions. A second predictive source decoder comprises, or is operably coupled to, logic arranged to obtain independent observations. The independent observations may be obtained by combining information input to the first predictive source decoder and an encoded signal of a predictor memory state of an encoder.

In one embodiment of the present invention, a communication unit comprises a receiver and logic operably coupled to the receiver for determining whether one or more errors have occurred in a received signal. A predictive source decoder is operably coupled to the logic and is capable of representing a received signal, wherein the predictive source decoder comprises or is operably coupled to logic arranged to bootstrap the predictor memory.

The provision of logic arranged to bootstrap the predictor memory, in error-prone channel conditions, provides the advantage of limiting the error propagation effects to the duration of the error event. As soon as an error event is over, the signal representing the predictor state may be used to bootstrap the predictor memory and from that point onwards the predictor may be allowed to operate as before.

In one embodiment of the present invention, the communication unit may comprise two scalar multipliers for receiving respective decoded values, wherein the two scalar multipliers are set in response to a signal sent -10 -by the encoder. The scalar multipliers may be set in response to an input signal applied to the encoder. The scalar multipliers may be applied to weighting logic to

provide multi-description coding gain.

In one embodiment of the present invention, a method of decoding in a communication unit comprises receiving a signal, decoding predictive source quantized parameters in the received signal; decoding quantized prediction memory information in the received signal; and determining whether one or more errors have occurred in a received signal. The method further comprises obtaining independent observations by combining information input to a first predictive source decoder (260) and an encoded signal of a predictor memory state of an encoder, when no errors are determined; or bootstrapping a predictor memory when one or more errors are determined.

Embodiments of the present invention address the need to provide such an error mitigation scheme where the source encoder employs predictive quantization, providing a means to prevent error propagation after detected error events, but also providing for an improvement in quality when all data is received error-free.

Referring now to FIG. 2, an overview of embodiments of the present invention is illustrated, highlighting both the encoder and decoder aspects. On the encoder side, the source signal 205 is input to a source predictive vector quantization (VQ) encoder 210, having memory 215, which is identical to 100 of Fig 1.

-11 -In accordance with embodiments of the present invention, both an input and an output of the source predictive VQ encoder 210 is also input to predictive VQ memory quantization logic 255. The predictive VQ memory quantization logic 255 comprises local source decoder logic 225, which comprises memory 230. Initially, the memory 230 is set to zero. An output from the local source decoder logic 225 is input to subtractor logic 235, which also receives the input signal 205, to produce an error signal. This error signal is input to logic 245 which is the functional inverse of the predictor employed in the predictive VQ encoder and decoder. The functional inverse of the predictor is performed in order to derive a signal that represents the ideal quantizer state, given the input signal to the quantizer and the quantization decision just made by the predictive quantizer logic.

The output of logic 245 represents the ideal value of the current state of the predictive VQ memory 215, subject to the decision(s) 1. It is this value which is quantized with VQ 250 and sent as additional embedded data I over the communications channel. The additional data can be embedded using any known combining logic, as well known

in the field of the present invention.

At the decoder, a main bitstream is received, along with additional data, in an embedded or scalable fashion. The main bit-stream is input to two parallel predictive VQ decoders 260, 275, which comprise respective PVQ memory elements 265, 280. A first predictive VQ decoder 260 has a conventional PVQ memory element, 265, which is driven by previous codebook decisions, as per the known decoder of FIG. 1. Notably, a second predictive VQ decoder -12 - 275 in an additional enhanced circuit 292 is provided, which has an LTP memory element, 280. Memory element 280 is preloaded using the quantized values of the additional embedded data 285 j, made by vector quantizer 250 in the encoder prior to applying the core codec decisions i.

It is noteworthy that the same VQ decisions i made in the conventional predictive VQ encoder 210 are applied to both of the predictive VQ decoders. The outputs from the two parallel predictive VQ decoders 260, 275 are input to weighted summation logic 270, which provides a high quality output when the communication channel is error free (in typical operating conditions) In accordance with one embodiment of the present invention, following an error where the conventional LTP memory 265 of predictive VQ decoder 260 would be corrupted, a bootstrap of LTP memory 265 may be performed. The bootstrap of LTP memory 265 may be performed using the additional enhancement data 285 j, which means that the outputs of the two predictive VQ decoders 260 and 275 will be identical for that sample or frame.

Whilst embodiments of the present invention are described with reference to a predictive VQ encoder and predictive VQ decoder, it is envisaged that the inventive concept is equally applicable to any encoder or decoder, where transmission errors may occur and predictive quantization is employed. Such schemes employ predictive quantization and additional bandwidth may be available to simultaneously provide greater error resilience, when -13 -channel errors are present. Furthermore, embedded coding schemes in such applications provide improved quality during error-free periods.

Thus, embodiments of the present invention allow the predictor state of a predictive quantizer to be coded for transmission. This allows error propagation to be minimized when channel errors are present. In addition, it also realizes a gain in performance, as provided by the additional data rate of the transmission, when there are no channel errors. One embodiment of the present invention achieves this dual goal by re-calculating the predictive quantizer predictor memory state, so that it may be independently quantized.

Embodiments of the present invention provide an improvement to the known principle of Multiple Description Coding (MDC), as described in V.K. Goyal, "Multiple Description Coding: Compression Meets the Network"; IEEE Signal Processing Magazine, Vol. 18, No. 5, pp74-93, September 2001. In known MDC techniques, two coded versions are transmitted over different transmission routes. Thereafter, one or both coded versions are employed at the decoder to yield a near optimal reproduction of the original source signal by optimizing the predictor in order to provide the MDC gain in the presence of errors of known statistics. Such a technique does achieve good performance near to the Rate Distortion bound when the channel conditions are known.

However, the technique can suffer error propagation problems when very high error rates are present.

-14 -However, in contrast to classical MDC coding, embodiments of the present invention deliberately incorporate asymmetry in the quality and data rate of the two MDC channels, such that the transmission can be sent via a single path with embedded additional data, and not sent via independent transmission paths. In effect, in contrast to known techniques, the predictor remains unchanged from a conventional predictive quantizer, and rather than the two similar independent transmissions of classical MDC, we have a single transmission path with deliberate asymmetry.

In summary, in the decoder, embodiments of the present invention propose to either: (i) obtain independent observations using both versions of the predictor memory state and combine them

to obtain a Multi-Description coding gain (in

substantially error free channel conditions), or (ii) Bootstrap the predictor memory (after detecting errors) Embodiments of the present invention involve quantizing a state of the predictor memory in such a way that, when it is combined with the predictive quantizer error value, it will provide an independent estimate of the input signal being quantized so that the quantization noise may be decorrelated as much as possible from the basic predictive quantizer output.

When channel errors occur, this allows the predictor memory to be bootstrapped prior to any future predictions. However, it also ensures that, without errors, the two independent estimates of the input signal -15 -may be combined by applying MDC techniques, thereby providing a single version of the input signal with lower error variance than either independent estimate.

Referring now to FIG. 3, a first order AR predictive scalar quantizer 300 adapted to support embodiments of the present invention is illustrated. Here, the index I is obtained from input signal 105 in the same manner as in FIG. 1. However, in accordance with embodiments of the present invention, and given the index 1120, a version of the state memory may be derived by subtracting (in subtract logic 305) the effective value represented by the index 1 120 from input signal 105.

This derivation of the state memory is then applied to computation logic 310 and a second quantizer 315. In one embodiment of the present invention, the resultant error signal is passed through an inverse of the predictor to yield an estimate of the ideal LTP state, which may be quantized directly to derive index j. Alternatively, in another embodiment of the present invention, if computation logic 310 of G' is complex, a codebook search may be initiated to establish the quantizer memory state that best models the second error signal. Index j 320 is also sent across the channel 150.

Such a procedure ensures that the error signals between the two representations are de-correlated, as much as possible, from one another. However, it is noteworthy that the two indices 1 120 and 1 320 relate to the same time instant, thus ensuring that an approximation to -16 -is possible when both are received, even following an error.

In one embodiment of the present invention, a derivation of a new estimate of the predictor state is obtained, by subtracting the effect of the original PVQ quantizer decision from the input signal before quantization of the LTP state. This leads to the two estimates of the input signal a, and a' having uncorrelated quantization noise; a situation which would not exist if the PVQ memory 140 were quantized directly.

Let us consider the case where no channel errors were experienced in receiving 1155, up until the current data is received. In this case, two versions of the input signal 170 and a' 370 are available at the receiver.

In the case where errors in the transmission of 1 120 are experienced, accompanying the next frame of data as determined with index 1155 at the receiver, and assuming it is error free, there will be a value of the predictor memory that may be used to bootstrap the predictor in order to derive an estimate of input signal 105 and ensure that when the next index 1155 is received, the predictor provides an output close to that of the encoder.

In one embodiment of the present invention, the values of the scalar multipliers g' and h' 380, 385 may be fixed over all time. Alternatively, in another embodiment, the values of the scalar multipliers g' and h' 380, 385 may be locally optimised over shorter periods of time and -17 -correspond to values sent from the encoder to the decoder. These values may also be computed with reference to the original input signal 105. It can be shown that the optimum values of g' and h' 380, 385, for the two gains at any particular time, are given by; I' -\2 / -\I,, -i _ui Isa-a) -Ia-a)a -a TI --gj --\2 I -V -\\ (I -\2 I -\I, a-a) -a-a)a -a+a -a) -a-afla -a)) [1] In this manner, embodiments of the present invention allow the predictor state of a predictive quantizer to be coded for transmission. This allows error propagation to be minimized when channel errors are present. It also allows a gain in performance to be realized, due to the additional data rate of the transmission when there are no channel errors. The key to achieving this dual goal is the re-calculation of the LTP memory state so that it may be independently quantized.

Referring now to FIG. 4, a flowchart 400 illustrates an auto-regressive (AR) predictive quantizer decoding operation according to embodiments of the present invention. The decoding operation commences with the decoder receiving a signal sent from the encoder, as described previously. The received signal is then decoded to obtain predictive source quantizer parameters, in step 405 and quantized prediction memory values, in step 410.

A determination is made, in step 415, as to whether previous data has been received in error. If it is determined in step 415 that the received signal comprises one or more errors, then the speech decoder is arranged -18 -to bootstrap both predictors' memory, as shown in step 420. Thereafter, both predictive vector quantized (PVQ) values from the memory values are then decoded in step 425. The process then moves to the next sample, as shown in step 445, and the process loops back to step 405.

Alternatively, if no errors are detected in step 415, then the second predictive VQ memory is bootstrapped in step 430. Both predictive vector quantized (PVQ) values from the memory values are then decoded in step 435. The predictive vector quantized (PVQ) values from the bootstrapped memories are then weighted and summed in step 440.

The process then moves to the next sample, as shown in step 445, and the process loops back to step 405.

Embodiments of the present invention offer particular benefits in embedded coding schemes, where predictive quantization is employed and additional bandwidth is available to simultaneously provide greater error resilience, when channel errors are determined as being present, and better quality during error-free periods.

In particular, it is envisaged that the aforementioned invenLive concept can be applied by a semiconductor manufacturer to any predictive encoding/decoding integrated circuit. It is further envisaged that, for example, a semiconductor manufacturer may employ the inventive concept in a design of a stand-alone device, or application-specific integrated circuit (ASIC) and/or any other sub-system element to support encoding/decoding.

-19 -It will be understood that the improved communication unit, integrated circuit and method of operation therefor, as described above, aims to provide at least one or more of the following advantages: (i) When operating in an error-prone transmission channel, the encoder-decoder arrangement limits the propagation of errors within the predictor.

(ii) When operating in a transmission channel where there are no errors, the encoder-decoder arrangement provides independent quantization of a source signal.

(iii) Advantageously, independent quantization of a source signal can be averaged with the original signal to provide a new signal with better fidelity (for example, lower quantization noise variance) (iv) The inventive concept facilitates separate coding of a predictor state of a predictive quantizer for transmission, through re-calculation of the ideal predictive quantizer memory state to allow it to be independently quantized. It will be appreciated that any suitable distribution of functionality

between different functional devices or elements or logic, may be used without detracting from the inventive concept herein described. Hence, references to specific functional devices or elements or logic are only to be seen as references to suitable means for providing the described functionality, rather than indicative of a strict logical or physical structure or organization.

Aspects of the invention may be implemented in any suitable form including hardware, software, firmware or -20 -any combination of these. The elements and components of an embodiment of the invention may be physically, functionally and logically implemented in any suitable way. Indeed, the functionality may be implemented in a single unit or IC, in a plurality of units or ICs or as part of other functional units.

Although the present invention has been described in connection with some embodiments, it is not intended to be limited to the specific form set forth herein.

Rather, the scope of the present invention is limited Only by the accompanying claims. Additionally, although a feature may appear to be described in connection with particular embodiments, one skilled in the art would recognize that various features of the described embodiments may be combined in accordance with the invention. In the claims, the term comprising' does not exclude the presence of other elements or steps.

Furthermore, although individual features may be included in different claims, these may possibly be advantageously combined, and the inclusion in different claims does not imply that a combination of features is not feasible and/or advantageous. Also, the inclusion of a feature in one category of claims does not imply a limitation to this category, but rather indicates that the feature is equally applicable to other claim categories, as appropriate.

Furthermore, the order of features in the claims does not imply any specific order in which the features must be performed and in particular the order of individual steps in a method claim does not imply that the steps must be -21 -performed in this order. Rather, the steps may be performed in any suitable order. In addition, singular references do not exclude a plurality. Thus, references to "a", "an", "first", "second", etc. do not preclude a plurality.

Thus, an improved communication unit and an integrated circuit and method of operation therefor have been described, wherein the aforementioned disadvantages with

prior art arrangements have been substantially

alleviated.

Claims

-22 -Claims 1. A communication unit (100) comprising a predictive

source encoder (210, 255) capable of representing an input signal, wherein the predictive source encoder (210, 255) comprises or is operably coupled to prediction memory (230), and quantization logic (250) capable of representing the state of the prediction memory (230) associated with the predictive source encoder (210, 255)
2. The communication unit of Claim 1 wherein the quantization logic (250) is operably coupled to subtractor logic (235) and derives an independent signal by subtracting an effect of one or more quantization decision(s) of the predictive source encoder from the input signal.

3. The communication unit of Claim 1 or Claim 2 wherein the independent signal passed through a function performing an inverse of the predictor provides a signal representing the predictor state to be quantized.

4. The communication unit of any preceding Claim further comprising combining logic to combine the predictive values of the input signal and the quantized memory state into a single embedded data stream for transmission.

5. A method of predictive source encoding in a communication unit, the method comprising: encoding an input signal, in a predictive source encoder, to produce a first encoded signal; -23 -quantizing prediction memory associated with the predictive source encoder; encoding the quantized prediction memory to produce a second encoded signal; and transmitting a signal comprising both the first encoded signal and the second encoded signal.

6. The method of Claim 5 further comprising: subtracting an effect of one or more decision(s) of the predictive source encoder from the input signal; deriving an independent signal in response thereto; and using an inverse of a predictor of the predictive source encoder Lo derive the prediction memory to be quantized.

7. The method of Claim 5 further comprising: subtracting an effect of decisions of the predictive source encoder from the input signal; deriving an independent signal in response thereto; and determining an optimum quantized value of prediction memory through an analysis by synthesis process.

8. The method of Claim 7 further comprising: minimizing an error, or perceptually weighted error, between a prediction from a quantized value of prediction memory and a derived independent signal.

9. The method of any of preceding Claims 5 to 8 further comprising: -24 -combining predictive values of the first encoded signal and second encoded signal into a single embedded signal for transmitting.

10. A communication unit comprising two predictive source decoders (260, 275) capable of representing a received signal, wherein a first predictive source decoder (260) derives a predictor memory state from previous decoding decisions, and wherein a second predictive source decoder (275) comprises or is operably coupled to logic arranged to obtain independent observations by combining information input to the first predictive source decoder (260) and an encoded signal of a predictor memory state of an encoder.

11. A communication unit comprising a receiver, logic operably coupled to the receiver for determining whether one or more errors have occurred in a received signal, and a predictive source decoder (260, 275) operably coupled to the logic and capable of representing a received signal, wherein the predictive source decoder (260, 275) comprises or is operably coupled to logic arranged to bootstrap the predictor memory.

12. The communication unit according to Claim 10 or Claim 11 further comprising logic arranged to determine whether one or more errors have been incorporated into a received signal, and in response to determining no errors: the logic obtains independent observations by combining information input to the first predictive source decoder (260) and an encoded signal of a predictor memory state of an encoder; -25 -or in response to determining one or more errors: the logic bootstraps the predictor memory.

13. The communication unit according to any of preceding Claims 10 to 12 further comprising two scalar multipliers (380, 385) for receiving respective decoded values, wherein the two scalar multipliers (380, 385) are set in response to a signal sent by the encoder.

14. The communication unit according to Claim 13 wherein the scalar multipliers (380, 385) are set in response to an input signal a (105) applied to the encoder.

15. The communication unit according to Claim 13 or Claim 14 wherein the scalar multipliers (380, 385) are applied to weighting logic (270) to provide multi-

description coding gain.

16. A method of decoding (400) in a communication unit comprising: receiving a signal, decoding (405) predictive source quantized parameters in the received signal; decoding (410) quantized prediction memory information in the received signal; determining (415) whether one or more errors have occurred in a received signal; and obtaining independent observations (430) by combining information input to a first predictive source decoder (260) and an encoded signal of a predictor memory state of an encoder, when no errors are determined; or -26 -bootstrapping (420) a predictor memory when one or more errors are determined.