18 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 17, NO. 1, JANUARY 1999 Voice over ATM Using AAL2 and Bit Dropping: Performance and Call Admission Control Kotikalapudi Sriram, Senior Member, IEEE, and Yung-Terng (Y.T.) Wang, Member, IEEE Abstract— Asynchronous transfer mode (ATM) adaptation layer 2 (AAL2) has been designed for efficient transport of voice, fax, and voiceband data (VBD) traffic over an ATM virtual circuit. The protocol helps achieve low latency and high bandwidth efficiency while applying suitable compression methods on voice/VBD/fax calls and silence elimination on voice calls. We analyze the performance and capacity of an ATM multiplexer based on AAL2 adaptation. We assume that embedded adaptive differential pulse code modulation (ADPCM) is used to compress voice, and silence elimination is used to achieve statistical multiplexing gain. The embedded ADPCM coding scheme allows selective dropping of less significant bits of voice during congestion in the ATM/AAL2 multiplexer. We compare the call capacities of voice multiplexers with and without bit dropping (BD). The performance models and results presented here are based on fairly general assumptions and can be used for traffic engineering and call admission control in land-line or wireless ATM systems for a variety of voice/voiceband compression algorithms. A generalized algorithm for call admission control is also described. Index Terms— ATM adaptation layer, ATM adaptation layer 2 (AAL2), call admission control, packetized voice, performance modeling and analysis, statistical multiplexing, traffic engineer- ing, voice and telephony over ATM (VTOA), wireless networks. I. INTRODUCTION I N this paper, we consider an asynchronous transfer mode (ATM) adaptation layer 2 (AAL2) system similar to that described in [1], but with the additional possibility of se- lectively dropping the less significant bits of voice packets during periods of congestion. Dropping less significant bits of voice for relieving congestion depends on the ability of the transport protocol to handle variable length packets, and is currently used in frame-relay networks. Because of the fixed ATM cell size, bit dropping (BD) on voice was not viable in ATM networks prior to the development of the AAL2 protocol. By allowing variable length packets within the ATM cell payload, the AAL2 protocol makes it possible to use BD. As described in [1] and [2], AAL2 also makes it possible to take advantage of voice silence periods. In this paper, we show that the combination of silence elimination and BD enables high overall statistical multiplexing gain and makes voice transmission over ATM very bandwidth efficient. It may be noted that the concept of selective discard of whole ATM cells containing less significant voice information Manuscript received February 1998; revised July 1998. This paper was presented in part at the IEEE ATM ’98 Workshop, Fairfax, VA, May 26–29, 1998. The authors are with Bell Laboratories–Lucent Technologies, Holmdel, NJ 07733-3030 USA. Publisher Item Identifier S 0733-8716(99)00007-4. was considered in the past [3], [4]. But these schemes entailed large packetization delays and had a coarse granularity in terms of separating bits of voice information into different orders of significance. The AAL2 protocol overcomes the shortcomings of such schemes by allowing variable length AAL2 packets within ATM cells. We present queuing analysis of AAL2 packet voice multi- plexers with and without BD drawing upon the models and insights presented earlier in [5]–[8]. The combined packet voice traffic from multiple voice sources tends to be very bursty due to correlations amongst successive interarrival times [5]. In our analytical models, the burstiness and cor- relations are well captured by a two-parameter approximation for a multiplexer without BD [5], [6]. In a voice multiplexer with BD, the effects of the burstiness/correlations are mitigated due to the traffic smoothing properties of BD. Consequently, a Poisson approximation for the combined packet voice traffic is quite valid for the multiplexer with BD, and an approximation works very well [8]. The in denotes state-dependent deterministic service time, and denotes the queue or buffer size in packets. Using the analytical models, we obtain several performance measures for both multiplexers (with and without BD), e.g., the mean, standard deviation and high percentile delays; packet loss ratio; mean bits per sample, etc. Based on the criterion of meeting certain requirements on the performance measures, we compute the effective bandwidth per voice call (as a function of the number of voice sources). The effective bandwidth is the minimum bandwidth that needs to be allocated to each voice call in order not to violate stated performance requirements, and it decreases as a function of the number of voice calls multiplexed. We make use of the effective bandwidth calculations to provide the multiplexer capaci- ties for various link bandwidth values. Further, we present algorithms for bandwidth allocation and deallocation when calls arrive and depart, respectively. These algorithms may be used as part of the overall call admission and bandwidth management scheme. The algorithms are general and may be used for a mix of voice, VBD, and fax calls in multiplexers with or without BD. II. AN AAL2 PACKET VOICE MULTIPLEXER WITH BD In this section, we describe the organization of an AAL2 voice packet amenable to BD. We also describe some method- ologies for implementing BD in an AAL2/ATM voice multi- plexer. 0733–8716/99$10.00  1999 IEEE