GB2346785A - Extending the resolution of a codebook - Google Patents

Abstract

A Communication system e.g. mobile radio using vector quantisation techniques for speech processing includes reference signals arranged in an original element list 70 of a codebook e.g. in a speech coding-decoding arrangement. The system utilises extra resources in the system in the form of extra data bits, free time slots, frequency channels or spreading codes for extending the resolution of the codebook. Based on the extra resources received, a space representing at least a first portion 72 of the original element list is divided into space representing at least two sub-lists 76. A full VQ (vector quantisation) codebook is progressively divided into sub-codebooks and a centroid, i.e. a fixed vector, is calculated for each sub-codebook from the centroid of the original list or codebook. Upon receiving an input signal a codebook index is allocated using a centroid of the original codebook together with a centroid of at least one extended codebook or sub-codebook to reference the optimum centroid for the received signal.

Description

1 2346785 SPEECH CODER FOR A COMMUNICATIONS SYSTEM AND METHOD FOR

OPERATION THEREOF

Field of the Invention

This invention relates to communication systems and more particularly to vector quantization techniques for speech processing (encoding/decoding) in radio communication systems.

Background of the Invention

Many voice communication systems, such as the TErrestrial Trunked RAdio (TETRA) system for private and public mobile radio users, use speech processing units to encode and decode speech. patterns. In such voice communication systems the speech encoder converts the analogue speech pattern from the radio microphone into a suitable digital format for transmission and the speech decoder converts a received digital speech signal into an appropriate analog speech pattern to be passed to the radio ear-piece.

As radio spectrum for such voice communications systems is a valuable resource, it is desirable to limit the channel bandwidth used, in order to ma3dmise the number of users per frequency band. Hence, the primary objective in the use of speech coding techniques is to reduce the occupied capacity of the speech patterns as much as possible, by use of speech compression techniques, without losing fidelity of the speech signal.

Traditional speech coding techniques have tended to use scalar quantization (SQ) techniques in processing the speech signal, due to their simplicity and good performance when the communication rate (i.e. bit rate) is sufficiently high. However, at low bit rates, SQ as a speech coding technique is not practical, as there is often less than 1 bit/sample available for encoding the speech signal. Therefore, to satisfy both the reduction in occupied capacity and yet retain a high degree of fidelity, more efficient and sophisticated quantization methods are required.

A popular solution in speech coding, in limited bandwidth communication systems, is the application of vector quantization (VQ). A prime incentive in using VQ can be found in Shannon's rate distortion theory, as known to those skilled in the art, which states that better performance can always be 5 achieved by coding vectors rather than scalars.

The process of vector quantization is to represent an input vector as a member of a set of fixed vectors. This set of fixed vectors is known as the VQ codebook. The fixed vector in the VQ codebook which best represents the input vector is found by exhaustively searching all members of the VQ codebook and selecting the fixed vector which gives the minimum distance measure (for example Euclidean distance) between it and the received input vector. This procedure requires that every fixed vector in the VQ codebook be searched in order to find the best representation of the input vector.

Although VQ has been shown to be very attractive and efficient in many areas of speech coding, it is not without its drawbacks, such as the comprehensive searching strategy to be used.

A popular method of reducing the complexity of a VQ codebook search has been to divide the full VQ codebook, of size W, into a number of VQ subcodebooks. This method effectively segments the vector space of the original VQ codebook into a number of fixed vector groups. A centroid is calculated for each of these sub-areas of vector space. The centroid is a fixed vector associated with each of the sub-areas whereby any received vector signal in the sub-areas is allocated the centroid fixed vector as being closest in vector space to the received vector signal. The centroid is calculated such that it provides the averaged minimum (often Euclidean) distance to each of the potential vector signals that are received.

A definition of Centroid can be found on page 351 of the book "Vector Quantization and Signal Compression" by Allen Gersho and Robert M. Gray.

A centroid is calculated for each group of W fixed vectors (where N may have a different value for each vector group) and each centroid is then used to categorise input vectors as being best quantized by a fixed vector in a particular VQ sub-codebook. When the input vector is received, a search algorithm performs a minimum distance calculation to identify the centroid to which the input vector is closest. The VQ sub-codebook of 'N' fixed vectors I corresponding to the chosen centroid is then searched to find the optimal fixed vector.

This selection is performed by choosing the fixed vector from the chosen VQ sub-codebook which gives the minimum distance between itself and the input vector. By searching through only the 'N' fixed vectors contained in the VQ sub-codebook, a fraction of the total number of fixed vectors is searched, as N<<M, providing a substantial computational saving.

Typically, radio communication systems use such speech coding techniques in order to transmit significant amount of speech information in a limited channel (or frequency bandwidth). Radio communication systems have traditionally been proprietary in nature, typically involving a single manufacturer, or a limited set of manufacturers, that provide infrastructure and other communication equipment.

Recently, there has been a move towards more open systems in which multiple manufacturers provide various components of the communication system. In such an open communication system, heterogeneous communication equipment must be inter-operable to be effective. To facilitate such inter-operability, standards are usually established and promoted which govern interface specifications, communication protocol, and the like. Strict adherence to such standards are ordinarily necessary in order to effectively operate an open communi cation system. With regard to speech coding technology, some standards have a bit-exact requirement, whereas others have a less rigid structure employing a floating point definition such as some Code Division Multiple Access (CDMA) communications systems.

In wireless radio frequency (RF) communication applications, one desirable standard is that of channel modulation. A channel modulation standard enables heterogeneous communication equipment to participate in radio communications. Once a particular channel modulation technique is chosen, this necessarily affects channel bandwidth, and consequently the quantity of information that may be transmitted over a given communication channel.

Depending on the application, trade-offs are often necessary when apportioning the channel bandwidth among data, error protection information, control information, and the like. For example, some applications, such as voice communication, may be real time in nature and therefore have different requirements than data communi cation that is not real time. Consequently, a channel modulation scheme is usually selected dependent upon the information likely to be communicated across a 5 communication channel.

Once a modulation scheme has been selected and accepted as a standard, communication equipment participating in the radio communications system must adhere to the standard in order to maintain compatibility and interoperability. This impacts the ability to provide enhanced communication capabilities through an increase in bandwidth, or other manipulation of communication parameters. Furthermore, once a speech coding technique has been selected for the standard, the performance of the speech coding is effectively fixed.

Although standards provide substantial benefits to the consumer through lower cost and greater options, standards inherently reflect compromises in technology. For specific applications, the established standards might not directly support certain performance needs. In such situations, advanced communication units may employ enhanced techniques that comply with the standard, and yet also allow for additional information to be transmitted, when conditions permit.

In such systems, the communications unit is defined as "backward compatible" whereby basic standard communications contained within the enhanced communication transmissions can be decoded by basic communications unit, and those units employing the enhanced techniques can decode the enhanced communication transmissions. Notably, a standard channel modulation scheme has a tendency to fix channel bandwidth which limits the opportunities to use additional information to improve communication. Yet, it is the standard channel modulation scheme that enables heterogeneous communication equipment to operate compatibly.

The performance of CELP based speech coders (e.g. in GSM and TETRA Standard) depends to a large extent on the size of the codebooks used. However, when using a standard compatible modulation scheme in the mobile radio system, the increasing size of the codebooks to improve the speech quality will result in the speech data stream not conforming with the standard.

I Thus a need exists to increase the size of a codebook, when communication parameters permit, and yet keep the speech data stream backward compatible to the basic codec defined in the standard. 5 Summary of the Invention

According to a first aspect of the present invention, a method for extending the resolution of a codebook in a communication system is provided. The method includes having a received signal that can be referenced to a first portion of an original element list of the codebook. The method comprises the steps of receiving extra resource in the communication system, and dividing at least the first portion of the original element list into at least two sub-lists, based on the extra resource, thereby extending the resolution of the codebook.

In this mannerY a standard codebook can be extended, when extra resource is made available, to better reflect the most appropriate vector of a received speech signal from a vector codebook structure. As a consequence, a speech signal can be generated by the speech coder that better represents the originally transmitted speech signal.

In a preferred embodiment of the invention, the step of dividing at least a first portion of the original element list includes determining a centroid of the original element list, dividing space representing the original element list into space representing at least two sub-lists and estimating centroids for each sub-list.

Furthermore, allocation of a centroid of a first sub-list as being the optimum centroid for the received signal is performed. Preferably the space representing the at least two sub-lists can be further divided dependent upon the extra resource made available. When extra resource is made available, a determination of the number of extra data bits available for extending the resolution of the codebook is made. This determination includes a determination of the number of sub-codebook levels to be generated using the extra data bits and a determination of the number of data bits to be allocated to each sub-codebook level. Centroids are determined for each of the sub-codebook levels.

In accordance with a second aspect of the present invention, a speech coder for operation in a speech communication system is provided. The speech coder has an original codebook and is capable of providing at least one extended codebook operably coupled to the original codebook upon receiving extra resource in the communication system. Preferably, the speech coder uses the original codebook for referencing received signals when extra resource is not available and uses the original codebook together with the at least one extended codebook when extra resource is available in the communication system. The extended codebook is generated by dividing the space represented by at least a first portion of the original codebook into space representing at least one extended codebook, based on the extra resource, thereby extending the resolution of the codebook. Preferably, the original codebook includes an original element list and the at least one extended codebook includes at least one extended element list, centroids of the at least one extended element list being determined from at least one centroid of the original element list. Upon receiving a signal a codebook index is allocated using a centroid of the original codebook together with a centroid of the at least one extended codebook, to reference the optimum centroid for the received signal.

In accordance with a third aspect of the present invention, a method of searching an extended codebook structure is provided. The method includes the steps of receiving an input signal determining a centroid in an original codebook structure which best reflects the received input signal, and determining further sub-centroids in the extended codebook structure, when the original codebook is extended, to further reflect the received input signal.

A preferred embodiment of the present invention will now be described, by way of example only, with reference to the drawings.

Brief Dgscrii2tion of the Drawings FIG. 1 shows a two-dimensional representation of a vector quantization (VQ) codebook.

FIG. 2 shows a two-dimensional representation of VQ sub-codebooks.

I FIG. 3 shows a two-dimensional representation of VQ sub-codebooks according to a preferred embodiment of the invention.

FIG. 4 shows a two-dimensional representation of a search path for an extended codebook in accordance with a preferred embodiment of the invention.

FIG. 5 is a flow chart detailing a method of generating the extended codebook.

Detailed Description of the Drawings

The trend in digital mobile radio systems is towards standardisation.

Hence, to provide performance enhancements over and above the standard system performance, an enhanced system has to conform with, or be backward compatible to, the basic standard. One example of the standardised performance in a speech communication system is that of the speech encoding/decoding operation to maintain a given speech quality in the communication system.

UK Patent Application UK 9523046.2 proposes means of increasing the data rate using a standard compatible modulation scheme and therefore increasing the bit rate available for coding the speech signal. However, a feature of the present invention is the use of this extra channel capacity to improve the accuracy of certain parts of the speech encoding process.

In CELP based coders, the major source of degraded speech quality lies in the vector quantization of the speech parameters (e.g. the LSF, excitation vectors, gain etc). Theoretically, it is better to have an infinite sized codebook for vector quantization to minimise errors. However, this is impractical. Therefore there is a trade-off between increasing the size of the codebook, to improve the performance of the codec towards the theoretical optimum, and saving memory space and computational complexity.

In order to achieve compatibility with the standard specification, an extended codebook structure, where indexing of the codebook is divided into at least two parts, is required. The first part provides an index of the vector chosen from the standard codebook; the speech data stream using only the first part therefore conforms to the standard and is transmitted to the decoder using the standard data channel. Only this data stream is used in any conformance tests. The second part is transmitted using the extra bit rate provided by the improved modulation scheme. This second speech data stream, when incorporated with the first part, provides the index of the vector chosen from the extended codebook.

Referring first to FIG. 1, a simple two-dimensional representation of a vector quantization (VQ) codebook 10 is shown. The VQ codebook comprises a number of fixed vectors 12.

In operation the VQ codebook 10 would typically comprise a number of dimensions having significantly more fixed vectors represented than shown in FIG. 1. The fixed vector in the VQ codebook 10 that best represents the input vector is found by exhaustively searching all fixed vectors of the VQ codebook 10 and selecting the fixed vector which gives the minimum distance measure between itself and the input vector.

Referring now to FIG. 2 the two-dimensional VQ codebook of FIG. 1 is shown, wherein the vector space of the VQ codebook 10 has been divided into a nunber of VQ sub-codebooks 22, 24, 26 and 28, as knovrn in the prior art.

By way of example only, the VQ codebook 10 has been divided into four VQ sub-codebooks 22, 24, 26 and 28 (segments) by applying the segment boundaries 14 and 16 to the original VQ codebook 10. Centroids 32, 34, 36 and 38 are then determined for each of the four segments 22, 24, 26 and 28 respectively.

In operation the centroids are used to categorise the input vectors as being best represented by a fixed vector in one of the VQ sub-codebooks 22, 24, 26 or 28. When an input vector is received, the centroid closest to the input vector is calculated. The VQ sub-codebook according to the chosen centroid is then searched to find the most optimal fixed vector. The selection of the most optimal fixed vector is accomplished by choosing the fixed vector from the VQ sub-codebook that gives the minimum distance between itself and the input vector. By searching through a VQ sub-codebook, a fraction of the I total number of fixed vectors are searched, providing a substantial computational saving.

Referring now to FIG. 3, an extended codebook is shown, in accordance with a preferred embodiment of the invention. When extra resource is available in a communication system, for example a higher data rate than is generally available in the standard communication system, as in the case provided by UK Patent Application UK 9523046.2, an extended VQ codebook 40 is generated.

The original VQ codebook space is divided into say, four additional subcodebook spaces 42, 44, 46 and 48 having respective centroids 52, 54, 56 and 58. An input vector (Vi) 62 which falls into a region (Ri) 60 will be vector quantised to centroid Ci 64. The index which indicates the location of Ci 64 conforms with the standard which defines the codebook. However, with the provision of an extended codebook in a second region 40, N extra bits are provided to improve the quantization of this codebook. As such, Ri for the second region can be further partitioned into 2 A N1 regions (where N1 ≤ N) and N1 bits are used to index the centroids of this first level partition. For example, if N1=2, then Ri can be partitioned into 4 more regions as shown the centroids Cij can be indexed as:

Cio = 00, Cil = 01, Ci2 = 10, Ci3 = 11 (1) Further partitioning can then be made with each sub-region further partitioned into 2 A N2 regions where N2≤(N-Nl) and the indexing scheme of the centroids Cijk of this second level partition is similar to that of Cij which depends on the value of N2. This process can be repeated for q levels so that:

N1 + N2 + N3 + + Nq = N (2) When the input vector Vi is assigned to Ci, Vi can then be further assigned to Cij and then to Cijk until Vi is assigned to CQjk... q). Then Vi is assigned to the extended codeword indexed as:

Ci Cii Cijk CQjk q) (3) In this manner, the accuracy of the centroids within an extended codebook can be increased in accordance with the extra resource made available. Extra resource can be made available under a variety of conditions, particularly in a radio communications system, when the communications traffic is low and hence "free" time slots, frequency channels, or spreading codes are available. Furthermore, extra resource can be made available when the signal strength conditions are favourable, allowing extra data to be transmitted, without significant prejudice to the received signal and the quality of the demodulated (recovered) data.

Although the preferred embodiment is described as a 2-dimensional VQ codebook, it is within the contemplation of the invention that the VQ codebook can have a much larger number of dimensions, a larger number of segment boundaries, each containing a finite number of fixed vectors. It is also within the contemplation of the invention that other sub-dividing of respective original codebooks can be dynamically performed to optimise the use of the additional resource.

Figure 4 shows the search path 68 through the extended codebook with input vector Vi. A standard codebook 70 (codeword) is shown having a set of input vectors. A fixed input vector 72 Vi is shown being closest to centroid 74 Ci. If a single extension codebook 76 is used, the codebook region having as a centroid 74 Ci, has a number of regions and associated sub-centroids to further categorise the fixed input vector 72 Vi. The associated sub-region and sub-centroid 78 Cij, thereby more accurately reflects the fixed input vector 72 Vi.

If a double extension codebook 80 is used, the single extension codebook region having as a sub-centroid 78 Ci, has a number of further subregions and associated further sub-centroids to more accurately reflect the fixed input vector 72 Vi. The further associated sub-region and subcentroid 82 Cijk, thereby more accurately reflects the fixed input vector 72 Vi. In this manner, the codebook extension process can be extended ad infinitum, to sub codebook 84 having the ultimate subcentroid 86 Cijk..q.

The extended codebook index 88 is shown having an increasing length of centroid indexing from Ci 90, through CiJ 92 and CiJk 94 to Cijk..q 96. The size of the extended codebook is in effect increased by a factor of I 2AN. This is a very efficient way of using the extra resource/bits provided because this scheme only needs one or more bits to operate. For every extra bit, the size of the codebook is doubled.

There are two reasons for not using all N bits in the first level of partition. First, different levels of protection can be applied to each dfferent partition level of the codeword index; e.g. Cij bits are heavily protected and C(ij... q) bits are least protected. Secondly, the encoder can transmit the extra speech data stream to the decoder in all channel conditions and no extra signalling to the decoder is needed. When burst errors occur in the extended codeword index, the decoder can still decode a useful codeword using the corrupted codeword index up to the bit just before an error occurs. If in the case of using all N bits in the first partition level, the corrupted extended codeword index becomes useless then the decoder can only use the standard codeword index Ci.

To generate the extra codeword not defined by the standard, a large amount of speech data from different speakers under different background conditions can be fed through the encoder thus ensuring the codebook is properly populated and represents a wide range of conditions. All input vectors assigned to a particular Ci can then be used to further partition the region Ri according to the value of N1 and the centroids of the sub-regions are stored as the first level partition codeword. Using the input vectors inside the sub-regions, each subregion can be further partitioned according to the value of N2 and the second level partition centroids are then stored. The process can then be repeated for q levels until the centroids of all level of partition are obtained.

FIG. 5 shows a flow chart detailing a method of generating the extended codebook. The process of generating an extended "tree- structure" VQ codebook starts at step 100. The number (N) of bits available in the data stream is determined, as shown in step 102, and the associated number (q) of tree levels required is determined, as in step 104. The bit allocation ratio (Nb) to each tree level is then determined, as shown in step 106, such that:

N1 + N2 + + Nq = N (4) The original centroids (Ci), used for the non-extended codebook, are then used as a basis for generating the extended centroids for the extended codebook. The tree level is also set to "0", as in step 108. VQ clustering using the tree structure VQ codebook, with levels up to the maximum tree 5 level, is then used to decide on the new centroid positions, as in step 110.

A large signal database is used in this process to accurately select the new centroid positions, as shown in step 112. A determination is then made as to whether there are enough data vectors for each of the extended centroid regions, as in step 114. If there are not enough data vectors, the clustering process of step 110 continues. If there are enough vectors in step 114, then for each extended centroid region at tree level, a design of a VQ codebook with a number of centroids (= 2Nb; where b tree level + 1) is made, as shown in step 116. The tree level value is then incremented, as in step 118.

If the tree level has failed to reach its maximum "q" value, in step 120, then the clustering process of step 110 is repeated. If the tree level has reached its maximum "q" value, in step 120, then all of the extended centroids of all of the tree levels are stored and the extended codebook design is complete, as shown in step 122.

Advantageously, the use of an extended codebook improves speech quality by more efficiently using extra resource available in the communication system, for example extra bit-rate provided by an improved modulation scheme in a mobile radio system. This improvement is achieved by micreasing the size of the codebooks used, and hence reducing vector quantization errors.

Furthermore, compatibility with the speech data stream with standard decoder is maintained so that a customer using a basic radio can still receive 11 standard" speech quality. This backward compatibility feature is particularly advantageous in communication systems where interoperability with other, more basic radios is a requirement.

The VQ codebook is, in the preferred embodiment, the TETRA algebraic code excited linear prediction (ACELP) codebook, as defined in document ETSI RES. 06.20 from the European Telecommunications Standard Institute, F0692 1, Sophia Antipolis, France. However, other speech codebook techniques would benefit from the inventive concept described herein.

I The present invention is useful and can be applied to any lossy source or data compression procedure which uses vector quantization. It is also useful in still image transmissions where the encoder sends out the full resolution image bit stream, which can be received by various decoders with different quality of service displaying the same image at different resolution, for example a mobile communications unit having an improved display unit can demodulate all of the data streams and display the full image, whereas a handset communications unit may only be able to demodulate a subset of the bit stream and therefore display the image at a lower resolution. 10 Thus, a method for increasing the resolution of a VQ codebook is provided. The method includes determining when communications conditions are such that extra resource is available, and using the extra resource to generate an extended VQ codebook. 15

Claims (18)

Claims

1. A method for extending a codebook resolution in a communication system wherein a received signal can be referenced to a first portion of an original element list of the codebook, the method comprising the steps of. receiving extra resource in the communication system; and dividing at least the first portion of the original element list into at least two sub-lists, based on the extra resource, thereby extending the codebook resolution.

2. The method of claim 1 wherein the step of dividing at least a first portion of the original element list includes: determining a centroid of the original element list; dividing a space representing the original element list into space representing at least two sub-lists; and. estimating centroids for the at least two sub-lisst.

3. The method of claim 2 further comprising a step of designating a centroid of a first sub-list as being the optimum centroid for the received signal.

4. The method of any one of the preceding claims, further comprising a step of further dividing the at least two sub-lists dependent upon the extra resource.

5. The method of any one of the preceding claims, wherein the step of receiving extra resource includes a step of determining a number of extra data bits available for extending the resolution of the codebook.

6. The method of claim 5, wherein the step of determining the number of extra data bits available includes a determination of a number of subcodebook levels to be generated using the extra data bits and a determination of a number of data bits to be allocated to each subcodebook level.

7. The method of any one of the preceding claims further comprising a step of determining centroids for each sub-codebook level.

I

8. The method of any one of the preceding claims wherein the received signal represents a fixed vector.

9. The method of any one of the preceding claims wherein the original 5 element list comprises regions of a vector quantization (VQ) codebook.

10. The method of any one of the preceding claims wherein the co. cation system is a speech communication system and/or the codebook is a TETRA ACELP speech codebook.

11. A speech coder for operation in a speech communication system, the speech coder having an original codebook and being capable of providing at least one extended codebook operably coupled to the original codebook upon receiving extra resource in the speech communication system.

12. A speech coder in accordance with claim 11, wherein the speech coder uses the original codebook for referencing received signals when extra resource is not available and uses the original codebook together with the at least one extended codebook when extra resource is available in the speech communication system.

13. The speech coder of claims 11 or 12, wherein the extended codebook is generated by dividing a space representing at least a first portion of the original codebook into a space representing at least one extended codebook, based on the extra resource, thereby extending the resolution of the codebook.

14. The speech coder of any one of preceding claims 11 to 13, wherein the original codebook includes an original element list and the at least one extended codebook includes at least one extended element list, centroids of the at least one extended element list being determined from at least one centroid of the original element list.

15. The speech coder of any one of preceding claims 11 to 14 wherein upon receiving a signal a codebook index is allocated using a centroid of the original codebook together with a centroid of the at least one extended codebook, to reference the optimum centroid for the received signal.

16. A method of searching an extended codebook structure comprising the steps of: receiving an input signal; determining a centroid in an original codebook structure which best reflects the received input signal; and determining further sub-centroids in the extended codebook structure, the extended codebook structure being an extension of the original codebook structure, to further reflect the received input signal.

17. A method of searching an extended codebook structure substantially as hereinbefore described with reference to, and/or as illustrated by FIG. 4 of the drawings.

18. A method for extending the resolution of a codebook in a speech communication system structure substantially as hereinbefore described with reference to, and/or as illustrated by, FIG. 5 of the drawings.

I I