GB2252018A - TDM digital speech signal coding - Google Patents

Abstract

Speech signals are coded on a frame-by-frame basis, the resuming information is coded by a convolutional coder 4 and the coded bits of each frame are distributed (6-22) amongst a plurality of time slots within a TDM structure comprising time slots of relatively shorter duration separated by intervals of relatively longer duration. The coded constraint length of the convolutional code is less than or equal to the number of time slots over which the coded bits of one frame are distributed, where the coded constraint length is the smallest number of bits that always includes all output bits which are a function of a given uncoded bit. A coding system using the signal is resistant to random and burst errors. <IMAGE>

Description

TDM Digital Speech Signals The present invention is concerned with the coding of digital signals, employing an error-correcting code, for insertion into a time-division multiplex signal.

Error-correcting codes are basically of two types.

In block coding, bits to be coded are assembled into blocks of bits and each block is then coded independently to produce an output block containing a larger number of bits. In convolutional coding, each bit to be coded (or each group of more than one bit) is coded by a convolutional coder to produce a larger number of bits; here however the coder has a memory of past events and the output depends not only on the current input bit(s) but also on the history of the input signal.

The actual coding operation is so chosen that, if the coded signal is received with one, or perhaps more, bits in error, a decoder can recover the correct bits.

Obviously there is a limit to the number of errors that can be tolerated, and their proximity. A block code is characterised by the number of errors per block that can be tolerated, whilst in the case of a convolutional code correct decoding is possible provided not more than the permitted number of errors occurs within a certain number of consecutive coded bits; this latter number is the smallest number of bits (referred to here as the "coded constraint length" Kc) which always includes all those output bits which are a function of any given uncoded bit. As definitions of constraint length vary, it should be noted that, in this document, the constraint length K is for a coder without feedback the number of serial coder delay stages plus one.Thus if the coder changes state every a data bits and produces a total of k output bits in this state (whence the code rate R = a/b) then the coded constraint length Rc = K.

A desirable feature of coding systems is that they should have not only a good resistance to randomlyoccurring errors, but also - since many types of transmission channel are subject to short periods of high error rate ("bursts") - have a good resistance to burst errors.

For this reason it has been proposed to employ interleaving in which the sequence of output bits is changed so that bits close together in the coded signal are spaced apart for transmission.

Necessarily this process involves the introduction of additional delays into the coding/decoding process.

Whilst for many digital transmission applications this is not important; for others, notably the transmission of speech, the amount of delay incurred is very significant and can represent a severe constraint on the parameters of the interleaving process. Also it is to be noted that many digital speech coding systems (e.g. LPC coding) involve partition of the input speech signal into frames each of which is processed to produce a set of parameters to be transmitted for that frame. Necessarily this results in a delay of at least one frame period (typically 20ms) in the coding/decoding process.

It has already been mentioned that we are concerned with time-division multiplex systems. One particular example is time division multiple access (TDMA) systems, where each user is allocated a series of relatively shorter multi-bit timeslots separated in time by relatively longer periods (to service other users); however the invention is applicable to other TDM systems fitting this description.

One example of a coding system of the type under discussion is that adopted for the proposed ETSI "GSM" standard for pan-European mobile radio (see, for example "Pan-European Cellular Radio", D.M. Balston, lEE Electronics & Communication Engineering Journal, Jan/Feb 1989). There, speech is initially coded to produce 260 bits for each 20ms speech frame (i.e. 13kbit/s).

Of these 260 bits, 182 (plus 3 CRC bits and 4 tailing bits) are coded by a 1/2 rate convolutional encoder having a coded constraint length of 10 and capable of correcting up to 3 errors within this constraint length to produce 378 coded bits, and the remaining 78 bits are unprotected - i.e. a total of 456 bits per frame.

The bits for one frame are carried in eight timeslots of a TDMA structure containing 0. 577ms databursts at 4. 615ms intervals. The convolutionally encoded bits are interleaved. The distribution of bits in the interleaving process is constrained however by the timeslot structure; thus it is easy to allocate eight bits to separate timeslots thereby ensuring a temporal separation of at least 5ms, but the ninth bit must necessarily be within .5ms of one of the preceding eight.

Thus - given the coded constraint length of 10 - of any 10 consecutive bits two pairs will each be located in the same timeslot and the probability of two errors within this length being caused by a single error burst is high.

This situation may of course be alleviated by increasing the number of timeslots used but (assuming the timeslot repetition rate is unchanged) this will (in this example) increase the system delay to an unacceptable level.

Conversely, the problem is exacerbated if the number of timeslots allocated is reduced; suppose that we wish to transmit a speech frame in only four timeslots at 10ms intervals. In that case, three bits within the coded constraint length of 10 can lie within the same timeslot and therefore the probability of a single error burst causing three errors and hence "using up" all the error correcting capacity of the code is high.

According to the present invention there is provided a method of coding speech comprising: (i) analysing speech to obtain, for each of successive time frames of speech, a plurality of information bits; (ii) coding the information bits using a convolutional code; (iii) distributing the coded bits of one frame amongst a plurality of time slots within a TDM structure comprising time slots of relatively shorter duration separated by intervals of relatively longer duration; wherein the coded constraint length of the code is less than or equal to the number of time slots over which the coded bits of one frame are distributed.

In another aspect, the invention provides an apparatus for coding speech comprising: (i) a speech coder for analysing speech to obtain, for each of successive time frames of speech, a plurality of information bits; (ii) a convolutional coder for coding the information bits using a convolutional code; (iii) means for distributing the coded bits of one frame amongst a plurality of time slots within a TDM structure comprising time slots of relatively shorter duration separated by intervals of relatively longer duration; wherein the coded constraint length of the code is less than or equal to the number of time slots over which the coded bits of one frame are distributed.

One embodiment of the present invention will now be described, by way of example, with reference to the accompanying drawings, in which: Figure 1 is a block diagram of a coding apparatus; and Figure 2 is a TDM timing diagram.

The apparatus shown in Figure 1 serves to receive, bit-serially at an input 1, digitally coded speech at 6. 75 kbit/s, namely 135 bits per 20ms speech frame. A 6. 75 kbit/s bit clock is received at an input 2. These 135 bits comprise 91 "protected" bits to be encoded using a convolutional code and 44 unprotected bits. After coding the bits of each frame are to be transmitted in four time slots of the TDMA frame structure shown in Figure 2. The slots are of duration 0. 577ms at intervals of approx toms. Each slot contains 114 bits of which 57 are used (the other 57 bits of the first two slots TS1, TS2 are used for the preceding frame and those of the second two slots are used for the following frame).

In order to alleviate the burst error problem discussed in the introduction, this apparatus employs a half rate convolutional code having a shorter constraint length than that employed in the GSM system discussed above, namely K=2, K =4. A penalty of employing a shorter code is that its error correcting capability is impaired, the specific coder to be described being capable of correcting 1 bit in 4 coded bits. This represents a capability of correcting up to 5 errors in 20 bits, compared with 6 for the coder discussed earlier. (Note that the maximum number of correctable errors depends on the relative positions of the erroneous bits within the constraint length). However, this disadvantage is more that offset by the benefits, as will be demonstrated below.

The 91 bits to be protected are conducted via a changeover switch 3 to a convolutional coder 4, of conventional construction, which implements the code discussed and produces for each input bit two bits D1, D2. A tailing unit 5 adds one tailing bit to the end of the 91 bit sequence and the convolutional coder thus produces (91+1)x2 = 184 bits. The output bits D1, D2, via a changeover switch 6, are clocked into four 46 bit serial in-parallel out shift registers 7, 8, 9, 10 to assemble bits for insertion (respectively) into four successive TDMA slots.The switch is toggled at half clock rate so that - if Dl(n), D2(n) are the coded bits for the nth data bit, then: Register 7 receives Dl(n), Dl(n+2), Dl(n+4) etc Register 8 receives D2(n), D2(n+2), D2(n+4) etc Register 9 receives Dl(n+1), Dl(n+3), Dl(n+5) etc Register 10 receives D2(n+1), D2(n+3), D2(n+5) etc So that any four "consecutive" coded bits appear in four separate registers.

When the 91 input data bits have been received at the input 1 the changeover switch 3 switches to conduct the 44 unprotected bits into a 44-bit SIPO shift register 11.

Each of the four groups of 46 bits from one of the registers 7-10 is assembled with a respective group of 11 bits from the register 11 to form a 57 bit group representing half a time slot. As these groups are created every 20ms and are to be transmitted in four time slots every 37. 5ms (taking into account overheads for control frames), each is concatenated with a 57 bit group from an earlier (or later) speech frame to form a 114-bit group. This is achieved by means of two banks of four 57 bit SIPO registers 12-15 and 16-19. The '57 bit groups are loaded alternately into the two banks by load pulses from a divide by two unit 20 driven by 20ms clock pulses.

The eight registers 12-19 are connected in pairs to form (for serial output) a 114 bit register clocked by respective pulses (at output data rate) + 2 3 4 from a timing unit 21. The timing of these pulses is indicated in Figure 2. The outputs for the four slots are combined in an Or-gate 22.

To obtain an accurate assessment of the improvement in error rate requires detailed probability analysis; however a significant improvement is obtained despite the use of a less powerful code. The results of tests on two systems using a (23,35) code and a (2,3) code are shown below giving the resident bit error rate RBER for difference values of carrier-to-interference ratio are given below in Table 1.

Table 1 C/I* 23,35 code RBER 2,3 code RBER 10dB 0. 496% 0. 277% 7dB 2. 2% 1. 17% 4dB 7. 12% 3. 91% * Typical range of cellular ratio carrier-tointerference ratio.

Claims (11)

1. A method of coding speech comprising: (i) analysing speech to obtain, for each of successive time frames of speech, a plurality of information bits; (ii)coding the information bits using a convolutional code; (iii) distributing the coded bits of one frame amongst a plurality of time slots within a TDM structure comprising time slots of relatively shorter duration separated by intervals of relatively longer duration; wherein the coded constraint length of the code is less than or equal to the number of time slots over which the coded bits of one frame are distributed.

2. A method according to claim 1 in which the interval between time slots is at least five times the duration of a time slot.

3. A method according to claim 1 in which the interval between time slots is at least ten times the duration of a time slot.

4. A method according to claim 1, 2 or 3 in which the said plurality of time slots fall within å period not exceeding 40ms.

5. A method according any one of claims 1 to 4 in which the coded constraint length of the code is equal to the number of time slots over which the coded bits of one frame are distributed.

6. An apparatus for coding speech comprising: (i) a speech coder for analysing speech to obtain, for each of successive time frames of speech, a plurality of information bits; (ii)a convolutional coder for coding the information bits using a convolutional code; (iii) means for distributing the coded bits of one frame amongst a plurality of time slots within a TDM structure comprising time slots of relatively shorter duration separated by intervals of relatively longer duration; wherein the coded constraint length of the code is less than or equal to the number of time slots over which the coded bits of one frame are distributed.

7. An apparatus according to claim 6 in which the interval between time slots is at least five times the duration of a time slot.

8. An apparatus according to claim 6 in which the interval between time slots is at least ten times the duration of a time slot.

9. An apparatus according to claim 6, 7 or 8 in which the said plurality of time slots fall within a period not exceeding 40ms.

10. An apparatus according any one of claims 6 to 9 in which the coded constraint length of the code is equal to the number of time slots over which the coded bits of one frame are distributed.

11. Apparatus for coding speech substantially as herein described with reference to the accompanying drawings.