ADPCM Equipment for 9.6-Kbps Data The ADPCM algorithm proposed by OKI Electric of Japan seems to be a formidable alternative for the standard. (an article taken from Telephony magazine, September 1987) [+] by Yoshihiko Yokoyama In 1982, the CCITT started work on developing a second digital encoding standard for speech, after decades of extensive use of PCM at 64 kbps in the A-law or u-law formats. The result of that effort was, the encoding standard of the 32-kbps ADPCM algorithm, known as CCITT recommendation G.721. It was recognized from the beginning that the algorithm should maintain adequate performance for voice-band data signals, although it was acknowledged that such signals were limited to data rates of up to 4800 bps for the state-of-the-art ADPCM algorithms. This has resulted in a virtual hesitation of widespread application of the standard in the public switched telephone networks (PSTNs), for which it was intended. Network operators have concluded that a fast-growing need exists for transmitting data at 9600 bps for their customers, and using G.721 makes that impossible. Susequently, the CCITT has embarked on a course of defining a digital encoding standard for digital circuit multiplication equipment (DCME), which combines time assignment speech interpolation (TASI) and a low-rate encoding technique such as ADPCM to form a very efficient means of transmitting speech. How to transmit data in such a system has been the subject of considerable debate and extensive effort by many experts in the field. It should be pointed out that, similar to the transcoding standard of G.721, interfacing with the DCME must be accomplished by means of an A-law or u-law encoded PCM signal format. The need for transmitting data up to 9600 bps has been recognized, and three algorithms have undergone scrutiny by a group of experts in the field. Two of the algorithms have the inherent capability of transmitting 9600-bps voice-band data at the 32-kbps rate, whereas the third algorithm under consideration is G.721, which does not have that capability. [+] PRESENT STANDARDIZATION EFFORTS DCME Aspects A DCME system is basically an all-digital implementation of the old concept of TASI. DCME systems operate on the statistical behavior of a group of talkers in a communication system. This is characterized by the average time that a talker on a connection is actually active, nominally assumed to be 35-40 percent of the total time the circuit is used for a call. Thus, the remaining time is available for time-interleaving the speech of other talkers. On the average, circuit usage can be increased or multiplied by a factor called digital speech interpolation (DSI) gain. Gain factors between 2 and 2.5 are commonly used, but these gain factors are dependent on the actual speech activity exhibited by the talkers. The larger the group of talkers, the more statistical stability is attained, and individual fluctuations in speech activity can be accommodated by the system. Long talk spurts by one talker are simultaneously compensated by silence or shourt spurts by another. Short durations of active speech, more than can be accommodated by available transmission capacity, do occur. Without "special means," this would result in what is known as clipped speech. In DCME, this special means is provided by instantly reducing the coding rate of one or more channels (talkers). That is, when the DCME operates nominally with ADPCM at 32 kbps during overload, this rate is reduced to 24 kbps for one or more channels. As sampling occurs at 8000 times per second, this means that the nominal channel being encoded at 4 bits/sample is reduced to encoding at 3 bits/sample during overload. This brings about a small degradation in performance by increased quantizing noise, but it occurs only sporadically due to the statistical nature of the phenomenon. Therefore, it is virtually imperceptible as long as the signal load to the DCME is strictly speech. When an appreciable part of the DCME load is data (more than 20 percent), special precaution must be taken to prevent noticeable degradation, because data signals do not exhibit the same on-off activity as speech. In fact, data are considered, generally, as being 100 percent active, thus providing no bearer circuit-sharing capability. When the DCME load is a mix of speech and data, it is clear overload will occur more often for the speech signals, resulting in an associated decrease in performance in the form of higher quantizing distortion. The choice of ADPCM algorithm for the DCME has an important bearing on this problem. [+] CCITT EFFORTS The CCITT is considering using the basic G.721 algorithm for speech at 32 kbps for DCME, but due to that algorithm's inablity to handle 9600-bps data at 32 kbps, encoding at 40 kbps per channel is needed for data signals at such rates. This is clearly having a more profound influence on the use of available bearer transmission capacity than if encoding of data could be limited to using the 32-kbps bearer rate per channel. For example, a 60-channel DCME system employing a proprietary ADPCM developed by OKI Electric of Japan can accommodate 10 percent data traffic up to 9.6 kbps, whereas G.721 ADPCM can only accommodate 6.7 percent data and maintain the same speech performance. Moreover, the DCME design is considerably simpler with the proprietary ADPCM, since there is no need to reconfigure the frame structure for including 5-bit/sample encoding for data. Another aspect of ADPCM in DCME systems is the need to tandem such systems for multilink networking purposes. It can generally be argued that no more than two DCME links should be allowed to be switched in any end-to-end connection. If such switching is performed by an analog switch (asynchronous tandeming), an accumulation of distortion will be experienced in the second link. However, if a digital switch would be employed, directly operating on the PCM output of the first DCME link, passing it digitally on to the second link (synchronous tandeming), no additional distortion will be experienced. Both the OKI ADPCM and the G.721-related technique in DCME application will have the "synchronous" capability as an inherent part of the design. A third algorithm, mentioned earlier, does, not possess that capability, and it will not be discussed. Digital switching will increasingly be employed in the public networks. Therefore, the loss of performance due to asynchronous tandeming, if it occurs at all, may only be temporarily experienced and should not pose a serious concern. This aspect of tandeming is not uniquely related to DMCE systems. Any application of 32 kbps could encounter the need for tandeming in a network. As digital switching will be increasingly applied, either by replacing analog switches or in new installations, the advantage of the ADPCM technique will be even more evident because of its capability of transmitting up to 9.6-kbps voice-band data signals. The CCITT nevertheless has decided to hold on to the G.721 technique, even though a clearly superior technique in now available. [+] OKI ADPCM PERFORMANCE Data Extensive performance measurements have been made in a carefully assembled test bed at COMSAT Laboratories. (This test bed received approval by the organizations that submitted ADPCM equipment for evaluation and comparison in a CCITT context. This made comparisons between algorithms valid and accurate.) The circuit in which the ADPCM equipment was tested included a simulated analog access link which introduced typical distortion effects (analog impairments) that a voice-band data signal may experience before being encoded by the ADPCM link. The typical performance after encoding by OKI ADPCM of a CCITT V.29 modem (The V.32 modems will perform even better than V.29 modems because of their inherent design. Thus, V.29 performance shown (graphs are not shown here in this text due to the inablility to draw or copy it here with this word processor) here is more critical to the user.) in terms of block error rate (BLER), as a function of S/N ratio of the data signal in the analog impairment circuit (i.e, just before being encoded), is illustrated in figure 1. Ther lower curve shown resulted after a single ADPCM encoding, whereas the higher curve resulted after a second ADPCM link was added to the first by means of an analog interconnection between the two links. Thus, this second curve is the result of asynchronous tandeming of two links. The curve showing single encoding perfomance applies also for the case of multiple encodings via digital switches, referred to as synchronous tandeming. A reference performance threshold of BLER = 10-2nd power at S/N =30.5 db (this reference point was selected by an SG XVIII expert group.) is well met by both curves. This indicates the excellent capability of the ADPCM equipment for transmitting 9.6-kbps V.29 signals. The performance of a V.29 modem operating at the back-off rate of 4.8-kbps tandem through four asynchronous encodings of the ADPCM equipment is shown in figure 2. For comparison, the dashed curve in fig. 2 shows the performance of the same modem when four asychronous links of G.721 ADPCM equipment are substituted for the OKI equipment. At S/N values to be expected in the networks, the OKI advanced ADPCM can perform two or more orders of magnitude better than G.721. This may not be required for this modem speed, but it is simply a consequence of its inherently more powerful predictor than that employed in G.721, and, as such, it provides an increased performance margin. Voice When considering ADPCM designs, the primary purpose has always been to provide high performance for voice signals. This objective has unquestionably been attained by the G.721 designers. Extensive subjective tests have proven the algorithm delivers the speech performance required for the networks. Similarly, the OKI ADPCM equipment provides the required performance when speech is transmitted through it. Tests similar to those used for evaluating the G.721 algorithm have been performed with the OKI ADPCM equipment, particulary for the English and Japanese languages. DCME Gain As has been pointed out earlier in the article, when applied in DCME systems, the proprietary ADPCM technique offers the advantage of encoding all voice-band data by using only only 4 bits/sample. This offers a bearer-channel efficiency advantage of up to 20 percent when transmitting 60 channels with 20 percent data. This includes a bearer-capacity increase to avoid speech degradation. Such an advantage may be particularly important for countries that may want to minimize their cost of communication. It should be emphasized, however, that without DCME, the main advantage of the propietary ADPCM resides in its capability of transmitting up to 9.6-kbps voice-band data. This has an important bearing on networks, since meeting this requirement is or will become indispensable. ------------------------------------------------------------- Yoshihiko Yokoyama is the General Representative for OKI America, Inc., New York office. --------------------------------------------------------------