GB2365297A

GB2365297A - Data modem compatible with speech codecs

Info

Publication number: GB2365297A
Application number: GB0018630A
Authority: GB
Inventors: David John Benjamin Pearce; Jon Alastair Gibbs; Julian Howard Richardson
Original assignee: Motorola Inc
Current assignee: Motorola Solutions Inc
Priority date: 2000-07-28
Filing date: 2000-07-28
Publication date: 2002-02-13
Also published as: GB0018630D0

Abstract

A data modem is compatible with multi-pulse linear predictive speech codecs. The modem comprises a modulator and a demodulator, wherein the modulator (fig 2) utilises a synthesiser of a multi-pulse linear predictive codec and the demodulator (fig 3) utilises the analyser of a linear predictive codec. The modem is then compatible with systems such as TETRA, ACELP or Full Rate GSM.

Description

<Desc/Clms Page number 1> DATA MODEM Field of the Invention This invention relates to a speech codec compatible data modem. In particular, but not exclusively, this invention relates to a multi-pulse linear predictive modem for use with systems such as TETRA and GSM.

Background of the Invention The voice channels of digital wireless systems make use of speech codecs which either render conventional data modems unusable or which greatly reduce the data rate obtainable over wireless voice channels. A modem which is compatible with the speech codec used in the voice channel allows higher data rates to be obtained than from modem techniques which have been designed for analogue telephone channels.

Many voice communications systems, such as the terrestrial trunked radio (TETRA) system for mobile radio users, and GSM, use speech processing units and algorithms to encode and decode speech patterns. These speech codecs are commonly multi-pulse linear predictive (LP) based codecs and these may be, for example, code excited linear prediction (CELP) speech coders, algebraic code excited liner prediction (ACELP) speech coders or vector sum excited linear prediction (VSELP) speech coders and these exhibit good performance at relatively low data rates. Such speech coders are well known and utilise linear prediction techniques to allow a coded representation of voice information to be transmitted instead of the original voice information or instead of a digitised version.

An ACELP codec essentially consists of three components; spectral

shaping coded in the form of quantised line spectral frequencies (LSF's typically ten LSF's every 30 ms frame; excitation pulses coming from the long term p redictor and pulse positions within four sub-frames (four pulses per sub- frame for TETRA, ACELP), and; quantised gains for each of the four sub-frames.

Accordingly, the speech coders derive an excitation signal by summing a long term prediction vector with one or more codebook vectors, with each vector being scaled by an appropriate gain prior to summing, and use a linear predictive filter to receive the resultant excitation vector and to introduce spectral shaping to produce a resultant synthetic speech. When properly configured, the synthetic speech provided by such a speak codec realistically mimics the original voice message.

Many conventional modem algorithms are known but none are presently known which are explicitly designed to be compatible with low data rate LP speech codecs such as ACELP codecs. Conventional modems suffer from the non-linearity that a speech coder introduces to a transmission channel. Summary of the Invention An object of the present invention is to provide an improved data modem. According to the present invention in a first aspect there is provided a data modem compatible with multi-pulse excited linear prediction based codecs.

According to the present invention in a second aspect, there is provided a data modem comprising a modulator and a demodulator, and. wherein the modulator utilises a synthesiser of a multi-pulse linear predictive codec and the demodulator utilises the analyser of a linear predictive codec,

In a preferred embodiment, the modulator is adapted to provide a periodic synchronisation pulse train and the demodulator is adapted to be able to synchronise with a signal from a modulator by using said synchronisation pulse train.

Brief Description of the Drawings Embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings, in which: Figure I shows the signal path through a wireless audio channel; Figure 2 shows a modulator using the structure of an ACELP codec synthesiser; Figure 3 shows a demodulator using the structure of the ACELP codecs analyser; and Figure 4 shows an ACELP codec.

Referring to Figure 1, a data modem for transmitting voice data over a voice channel has to transfer information via the voice channel parameters as an intermediate parameterisation. Figure I shows the signal path through the channel. The signal input from a modulator I is analysed by a codec encoder (or analyser) 2 as quantised coded parameters with error protection 3. This signal is then decoded at the receiver (CELP decoder) 4 to an analogue signal before being passed to a demodulator 5.

In embodiments of the invention, the modem may be in the form of a low cost digital signal processor (DSP) and appropriate interface circuitry or may be configured as a software product which executes on the host processor of a computer involved in data transfer. For this, the computer will need to have an analogue output and this may be obtained, for example, from a PCMCIA card modem or an analogue interface card (e.g. sound card). The interface to the radio

communications network may be via acoustic coupling for example or through a connector directly to a radio system's audio input (microphone) and output (loudspeaker), or may use other means.

The invention will be described further in relation to the specific case of the multi-pulse LP codec, used in the TETRA standard ACELP at 4.56 kbit per second. However, it should be appreciated that the invention has wider applicability than this and may be applied to channels employing other multi- pulse LP codecs, including GSM.

In the example of Figure 1, the encoded information used by the ACELP codec consists of- 1 . Spectral shaping coded in the form of quantised LSF's (e.g. 10 LSF's every 30 ins frame); 2. Excitation pulses coming from the long term predictor and pulse positions within four sub-frames (four pulses for sub-frame for TETRA ACELP); 3. Quantised gains for each of the four sub-frames.

In an important feature of embodiments of the invention, the codec parameterisation. uses a variant of a multi-pulse LP codec synthesiser (i.e. decoder) as part of the modulator and the codec's analyser (or encoder) as part of the demodulator.

Figure 2 shows schematically a modulator using the structure of the ACELP codec synthesiser. It will be recalled that the ACELP system basically splits speech signals into components representative of slow spectral changes and an excitation component. For the excitation component, PPM (pulse position modulation) may be used in embodiments of the invention. PPM has not generally been used in modem transmission up to now.

Referring to Figure 2, PPM data is split into a number of sub-frames 6#, 6b, 6, .... 6, One of the available pulses in one of the sub-frames 6,, is used as a synchronisation pulse. This enables the demodulator part to be synchronised with the modulator, as will be described in further detail below. The pulse positions in the other sub-frames are used to transmit information bits and their positions are recovered relative to the synchronisation pulses. The information bits and the various sub-frames are multiplied at a multiplier 7 with respective codebook sub- frame gains of the TETRA ACELP speech codebook. An example of this is defined in document ETSIRES.06.20 from the European Telecommunications Standard Institute, F06921, Sophia Antipolis, France. However, other speech codebook techniques may be used with embodiments of the present invention.

The output from multiplier 7 is applied to an LPC filter 8 where LSF values I to 10 (line spectral frequencies) are applied and the output of this is an analogue output 9. This output may then be used as the output from the modulator part of the modem.

All of the following modulation techniques are compatible with multi- pulse LP type codecs: Pulse position modulation (PPM) Pulse amplitude modulation (PAM) Pulse shaping modulation (PSM) Figure 3 shows how by using the codec analyser as part of the demodulator, the various types of data (PPM, PAM, PSM) may be obtained. Referring to Figure 3, the modulator receives a signal input at 10. The short term spectrum is extracted at a short term spectrum extractor 11. Note that each of the steps may be embodied in hardware or in software and the invention applies equally to these and to a combination of hardware and software. The extracted short term spectrum is applied to a PSM decoder 12 and PSM data 13

may be obtained, representative of LSFs. The extracted short term spectrum data I I is also applied to an inverse filter 14 which uses the input and the extracted short term spectrum to generate an output to a PPM decoder 15, The sync pulses (6,,) from the modulator are also detected at 16 and these are used in the PPM decoder in order to properly synchronise this. PPM data can be extracted at 17 after the PPM decoder and, by extracting the gains G I ... GN+1 at gain extractor 18, PAM data is output, representative of the gains of the sub-frames.

In order to make use of the bits available in the excitation pulses, the demodulator has to be synchronised with the modulator. This is done by using the regular train of fixed timing pulses in one of the sub-frames to provide a synchronisation pulse. This is a synchronisation sub-frame 6a shown in Figure 2. The pulse positions in the other three sub-frames are used to transmit information bits and their positions are recovered relative to the synchronisation pulses.

As discussed, the other bits used to carry data are the LSFs (line spectral frequencies) (PSM) and the gains of the sub-frames (PAM).

It is preferable, in some embodiments, in order to aid analysis, to switch off the long terni predictor of the codec in both the modulator and demodulator. Figure 4 shows a typical ACELP codec. Referring to the Figure, a signal input at 20 is applied to a short terni spectrum extractor 2 1. After being Q- switched 22, LSF values (PSM data) may be output at 23.

The analyser is generally unable to reliably recover every possible combination of quantised synthesis driving parameters. In particular, adjacently quantised bits are easily confused. Two possible approaches for achieving the maximum bit rate are as follows: I . Not to make use of adjacent quantised parameters but only to

encode those which a demodulator (based on the multi-pulse LP encoder) can reliably extract at an appropriate error rate.

2. To make use of all the quantisation parameters but to add additional error correction bits which correct mistakes made by the analysis of the demodulator.

Figure 4 shows one method of error correction.

Referring again to the Figure, index signals can be applied to an ACELP codebook 24. These are multiplied at a multiplier 25 with a first gain signal G1 and applied to an adder 26. Pitch lag signals 27 are applied to an adaptive codebook 28 and the resulting code is applied to a further multiplier 29 where it is multiplied with a second gain signal G2. This is also applied to the adder 26 and the resultant from the adder is applied to a perceptual weighting filter 30. The resultant is also applied as a feedback signal to the adaptive codebook 28 in order to provide the adaptive nature of this. An output signal from the perceptual waiting filter 30 is applied to a target 31 in the form of a subtractor which also receives a signal from a perceptual waiting filter 31 and an inverse filter 32 connected to the input as shown. The resultant of the subtractor function at the target 33 is an error signal 34.

In Figure 4, the dashed line 35 illustrates the error minimisation components.

In one non-limiting example, the likely bit utilisation for TETRA ACELP is estimated as follows:

Used reliably (bits per 30ms frame) Line spectral frequencies (LSF's): 2.5 Excitation pulses: 12 Sub-frame gains: 7 Total 21.5 This is based on the following: 32 (5 bits) filter shapes for the LSF's switched every 60ms [5/2 per frame]. One pulse per sub-frame with 16 (4 bits) possible location per sub-frame (3 sub-frames available per frame) [4 x 3 = 12 per frame].

4 gain level for each of the 3 PPM data sub-frames + 2 gain levels for the synchronisation sub-frame [3 x 2 + I = 7 per frame].

This provides a data rata of 710 bits per second. Final bit allocations and data rate achievable depends on the detailed implementation based on experimental optimisation through simulation.

In addition half-rate Trellis coded LSB gains could provide an extra 0.5 bits per sub-frame.

In general, the data rates which may be expected from such a modem over the TETRA ACELP channel are in the region of 5 00 to 1000 bits per second. However, data rates greater than this may indeed be possible.

Embodiments of the present invention may be used either to supplement existing data services which are integrated with voice, such as short data

messaging in the TETRA standard, or may be used to provide a point to point data service for infrastructures not equipped for circuit or packet mode data.

Where a modem must communicate over several different speech codec links comprising different speech codec characteristics, e.g. TETRA (30 ms frames) and GSM (20 ms frames) end-to-end modem negotiation and sounding or channel measurement, as employed in conventional voice band telephone modems, may be used in order to optimise the modulation parameters. Mechanisms for this are well known to the skilled reader.

Claims

CLAIMS 1. A data modem compatible with multi-pulse linear predictive speech codecs.
2. A data modem comprising a modulator and a demodulator, wherein the modulator utilises a synthesiser of a multi-pulse linear predictive codec and the demodulator utilises the analyser of a multi-pulse linear predictive codec.
3. A data modem as claimed in Claim 2, wherein in the modulator PPM data in the form of a plurality of sub-frames is applied to a multiplier where sub-frame gains are applied, before the application of LSF filtening, to generate an analog output.
4. A data modem as claimed in Claim 2 or Claim 3, wherein the analyser comprises means for extracting PSM, PPM and/or PAM data.
5. A data modem as claimed in Claim 4, comprising a short term spectrum extractor and a PSM decoder.
6. A data modem as claimed in Claim 4 or Claim 5, comprising an inverse filter and a PPM decoder.
7. A data modem as claimed in Claim 6, comprising a gain extractor, receiving an input from the PPM decoder to thereby generate PAM data.
8. A data modem as claimed in any preceding claim, comprising a modulator and a demodulator, wherein, in the modulator, data is provided in the form of a plurality of sub-frames, whereby data in one of the sub-frames is a synchronisation pulse.

<Desc/Clms Page number 11>
9. A data modem as claimed in Claim 8, wherein the other sub-frames convey information bits encoded in the position of speech codec excitation pulses which are recoverable relative to the synchronisation pulses.
10. A data modem as claimed in Claim 8 or 9, when dependent upon Claim 6 or Claim 7, wherein the demodulator comprises a synchronisation pulse detector.
11. A data modem as claimed in any preceding claim, including error correction means operable to introduce additional error correction bits.
12. A data modem as claimed in any preceding claim, wherein the codec is an ACELP codec.
13. A data modem as claimed in Claim 13, wherein the long-term predictor of the codee is not used.
14. A data modem as claimed in any preceding claim, wherein the parameters of the modulation scheme are adjusted as a result of end-to-end negotiation or channel measurement.
15. A data modem substantially as hereinbefore described with reference to, and as illustrated by, the accompanying drawings.
16. Use of a data modem as claimed in any preceding claim to transmit data over a wireless voice channel.