ADAPTIVE JOINT-DETECTION COMA VIDEO TRANSCEIVER
4 1 Background of the Invention
s The invention relates to burst-by-burst adaptive joint-detection Code Division Multiple Access (CDMA) s based transmission of multimedia signals, such as interactive video or audio, speech etc. 7 In contrast to the burst-by-burst reconfigurable CDMA multimedia transceivers described in this doc- β ument, the term statically reconfigurable found in this context in the literature refers to multimedia s transceivers that cannot be near-instantaneously reconfigured. More explicitly, the previously proposed ιo statically reconfigurable video transceivers were reconfigured on a long-term basis under the base stall tion's control, invoicing for example in the central cell region - where benign channel conditions prevail i2 - a less robust, but high-throughput modulation mode, such as 4 bit symbol Quadrature Amplitude Mod- i3 ulation (16QAM), which was capable of transmitting a quadruple number of bits and hence ensured a i4 better video quality. By contrast, a robust, but low-throughput modulation mode, such as 1 bit/symbol is Binary Phase Shift Keying (BPSK) can be employed near the edge of the propagation ceil, where hostile i6 propagation conditions prevail. This prevented a premature hand-over at the cost of a reduced video i7 quality. is The philosophy of the fixed, but programable-rate proprietary video codecs and statically reconfigurable i9 multi-mode video transceivers presented by Streit et at. for example in References [1] was that irrespec-
20 tive of the video motion activity experienced, the specially designed video codecs generated a constant
2i number of bits per video frame. For example, for videophony over the second-generation Global System
22 of Mobile Communications known as the GSM system at 13 kbps and assuming a video scanning rate of
23 10 frames/s, 1300 bits per video frame have to be generated. Specifically, two families of video codecs
24 were designed, one refraining from using error-sensitive run-length coding techniques and exhibiting the
25 highest possible error resilience and another, aiming for the highest possible compression ratio. This
26 fixed-rate approach had the advantage of requiring no adaptive feedback controlled bitrate fluctuation
2 smoothing buffering and hence exhibited no objectionable video latency or delay. Furthermore, these
28 video codecs were amenable to video telephony over fixed-rate second-generation mobile radio systems,
29 such as the GSM. 0 The fixed bitrate of the above proprietary video codecs is in contrast to existing standard video codecs,
i such as the Motion Pictures Expert Group codecs known as MPEG 1 and MPEG2 or the ITU's H 263 2 codec, where the time-vanant video motion activity and the vaπable-length coding techniques employed
33 result in a time-vaπant bitrate fluctuation and a near-constant perceptual video quality This time-vaπant
34 bitrate fluctuation can be mitigated by employing adaptive feed-back controlled buffeπng, which po- 5 tentially increases the latency or delay ot the codec and hence it is often objectionable for example in 3β interactive videophony The schemes presented by Streit et at in References [1] result in slightly vaπable 7 video quality at a constant bitrate, while refraining from employing buffeπng, which again, would result β in latency in interactive videophony A range of techniques, which can be invoked, m order to render the
39 family of vaπable-length coded, highly bandwidth-efficient, but potentially error-sensitive class of stan-
40 dard video codecs, such as the H 263 arrangement, amenable to error-resilient, low-latency interactive 4i wireless multimode videophony was summonsed in [2] The adaptive video rate control and packetisa- 42 tion algoπthm of [2] generates the required number of bits for the burst-by-burst adaptive transceiver, « depending the on the capacity of the current packet, as determined by the current modem mode Fur- ther error-resilient H 263-based schemes were contπved for example by Farber, Stembach and Girod
45 at Erlangen University [3], while Sadka, Eryurtlu and Kondoz [4] from Surrey University proposed a
46 range of improvements to the H 263 scheme Following the above portrayal of the pπor art in both video
47 compression and statically reconfigurable narroband modulation, let us now consider the philosophy of 4β wideband burst-by-burst adaptive quadrature amplitude modulation (AQAM) in more depth
49 In burst-by-burst adaptive modulation a higher-order modulation scheme is invoked, when the channel so is favourable, in order to increase the system's bits per symbol capacity and conversely, a more robust si lower order modulation scheme is employed, when the channel exhibits infeπor channel quality, in order
52 to improve the mean Bit Error Ratio (BER) performance. A practical scenaπo, where adaptive modula- j tion can be applied is, when a reliable, low-delay feedback path is created between the transmitter and
5 receiver, for example by supeπmposing the estimated channel quality perceived by the receiver on the
55 reverse-direction messages of a duplex interactive channel The transmitter then adjusts its modem mode
56 according to this perceived channel quality
57 Recent developments m adaptive modulation over a narrow-band channel environment have been pise oneered by Webb and Steele [5], where the modulation adaptation was utilized in a Digital European 59 Cordless Telephone - like (DECT) system The concept of vaπable rate adaptive modulation was also so advanced by Sampei et al [6], showing promising advantages, when compared to fixed modulation in si terms of spectral efficiency, BER performance and robustness against channel delay spread In another
62 paper, the numeπcal upper bound performance of adaptive modulation in a slow Rayleigh flat-fading
63 channel was evaluated by Torrance et al[l] and subsequently, the optimization of the switching threshold
levels using Powell minimization was used in order to achieve a targeted performance [S, 9]. In addition.
_5 adaptive modulation was also studied in conjunction with channel coding and power control techniques
66 by Matsuoka et at [6] as well as Goldsmith et at [10].
67 In the narrow-band channel environment, the quality of the channel was determined by the short term 6β Signal to Noise Ratio (SNR) of the received burst, which was then used as a criterion in order to choose
69 the appropriate modulation mode for the transmitter, based on a list of switching threshold levels, /„ [5, 9].
70 However, in a wideband environment, this criterion is not an accurate measure for judging the quality of i the channel, where the existence of multi-path components produces not only power attenuation of the
72 transmission burst, but also intersymbol interference. Subsequently, a new criterion has to be defined to
73 estimate the wideband channel quality in order to choose the appropriate modulation scheme.
4 2 Summary of the Invention
75 Particular and preferred aspects of the invention are set out in the accompanying independent and depen-
76 dent claims. Features of the dependent claims may be combined with those of the independent claims as
77 appropriate and used in combinations other than those explicitly set out in the claims.
78 The performance benefits of burst-by-burst adaptive modulation assisted CDMA are described, employ-
79 ing a higher-order modulation mode in transmission bursts, when the instantaneous channel quality is so favourable, ie when the received signal is unimpaired by co-channel interferers. This procedure is em- 8i ployed, in order to increase the system's bits per symbol (BPS) capacity and conversely, invoking a more β2 robust, lower order modulation mode, when the channel exhibits inferior channel quality. Therefore the as associated bit rate will be time-variant. i It is shown that due to the described adaptive modem mode switching regime a seamless multimedia as source-signal representation quality - such as video or audio quality - versus channel quality relation-
Bβ ship can be established, resulting in a near-unimpaired multimedia source-signal quality right across a? the operating channel Signal-to-Noise Ratio (SNR) range. The main advantage of the described tech-
8β nique is that irrespective of the prevailing channel conditions, the transceiver achieves always the best eg possible source-signal representation quality - such as video or audio quality - by automatically adjust-
9o ing the achievable bitrate and the associated multimedia source-signal representation quality in order to
9i match the channel quality experienced. This can achieved on a near-instantaneous basis under given
92 propagation conditions in order to cater for the effects of path-loss, fast-fading, slow-fading, dispersion,
93 co-channel interference, etc. Furthermore, when a mobile is roaming in a hostile out-doors - or even
94 hilly terrain - propagation environment, typically low-order, low-rate modem modes are invoked, while
95 in benign indoor environments predominantly the high-rate, high source-signal representation quality
96 modes are employed.
97 3 Brief Description of the Drawings
98 For a better understanding of the invention and to show how the same may be carried into effect reference
99 is now made by way of example to the accompanying drawings, in which:
1 List of Figures
ιoι 1 Signalling scenarios in adaptive modems 19
.2 2 Reconfigurable transceiver schematic 20
.03 3 Transmission burst structure of the FMA1 spread speech/data mode 2 of the FRAMES to* proposalfl l] 20 ιo5 4 Normalized channel impulse response for the COST 207 seven-path Bad Urban channel. 21 ιoβ 5 BER versus channel SNR 4QAM performance using BCH coded 13.7Kbps video, com- ιo7 paring the performance of matched filtering and joint detection for 2-8 users 21 ice 6 BER versus channel SNR 4QAM performance using turbo-coded 11.1Kbps video, com- io9 paring the performance of matched filtering and joint detection for 2-8 users 22 no 7 Video packet loss ratio versus channel SNR for the turbo-coded 11.1 Kbps video stream, in comparing the performance of matched filtering and joint detection for 2-8 users 22 ιi2 8 Video packet loss ratio (PLR) and feedback error ratio (FBER) versus channel SNR for us the three modulation schemes of the 2-user multi-mode system using joint detection. . . 23 ιi4 9 Decoded video quality (PSNR) versus channel SNR for the modulation modes of BPSK, us 4QAM and 16QAM supporting 2-users with the aid of joint detection. These results ιιβ were recorded for the Miss-America video sequence at SQCIF resolution (128x96 pels). 23 it? 10 Decoded video quality (PSNR) versus video packet loss ratio for the modulation modes us of BPSK, 4QAM and 16QAM, supporting 2-users with the aid of joint detection. The ιi9 results were recorded for the Miss-America video sequence at SQCIF resolution (128x96
i2i 1 1 Example of modem mode switching in a dynamically reconfigured burst-by-burst o-
122 dem in operation, where the modulation mode switching is based upon the SINR estimate
123 at the output of the joint-detector. 24
PDF ot the vaπous adaptive modem modes versus channel SNR 25 Throughput bitrate versus channel SNR compaπsion ot the three fixed modulation modes (BPSK, 4QAM, 16QAM) and the adaptive burst-by-burst modem (AQAM), both supporting 2 users with the aid of joint detection 25 Average decoded video quality (PSNR) versus channel SNR compaπsion of the fixed modulation modes of BPSK, 4QAM and 16QAM, and the burst-by-burst adaptive modem Both supporting 2-users with the aid of joint detection These results were recorded for the Miss-Ameπca video sequence at SQCIF resolution (128x96 pels) 26 Video packet loss ratio (PLR) versus channel SNR for the three modulation schemes ot the multi-mode system, compared to the burst-by-burst adaptive modem Both systems substain 2-users using joint detection 26 Decoded video quality (PSNR) versus video packet loss ratio compaπsion of the fixed modulation modes of BPSK, 4QAM and 16QAM, and the burst-by-burst adaptive modem Both supporting 2-users with the aid of joint detection These results were recorded for the Miss-Ameπca video sequence at SQCIF resolution (128x96 pels) 27
,3 3.1 State-of-the-art i o Burst-by-burst adaptive quadrature amplitude modulation (AQAM) was contrived by Steele and Webb [5],
MI in order for the transceiver to cope with the time-variant channel quality of narrowband fading channels. i42 Further related research was conducted at the University of Osaka by Sampei and his colleagues, investi-
1 3 gating variable coding rate concatenated coded schemes [6], at the University of Stanford by Goldsmith
14 and her team, studying the effects of variable-rate, variable-power arrangements [10] and at Southamp-
145 ton University in the UK, investigating a variety of practical aspects of AQAM [12, 13]. The channel's
146 quality is estimated on a burst-by-burst basis and the most appropriate modulation mode is selected in or- i4 der to maintain the required target bit error rate (BER) performance, whilst maximizing the system's Bit us Per Symbol (BPS) throughput. Using this reconfiguration regime the distribution of channel errors be-
»9 comes typically less bursty, than in conjunction with non-adaptive modems, which potentially increases
.JO the channel coding gains. Furthermore, the soft-decision channel codec metrics can be also invoked in
151 estimating the instantaneous channel quality, irrespective of the type of channel impairments. ιs2 A range of coded AQAM schemes were analysed by Matsuoka et al [6], Lau et al [14] and Gold-
15 smith et al [10]. For data transmission systems, which do not necessarily require a low transmission
154 delay, variable-throughput adaptive schemes can be devised, which operate efficiently in conjunction ιs5 with powerful error correction codecs, such as long block length turbo codes. However, the acceptable
156 turbo interleaving delay is rather low in the context of low-delay interactive speech. Video communica-
157 tions systems typically require a higher bitrate than speech systems and hence they can afford a higher isβ interleaving delay.
159 The above principles - which were typically investigated in the context of narrowband modems - were i6o further advanced in conjunction with wideband modems, employing powerful block turbo coded wide- i6i band Decision Feedback Equaliser (DFE) assisted AQAM transceivers [15]. A neural-network Radial
162 Basis Function (RBF) DFE based AQAM modem design was proposed in [16], where the RBF DFE
163 provided the channel quality estimates for the modem mode switching regime. This modem was capa- ,64 ble of removing the residual BER of conventional DFEs, when linearly non-separable received phasor i65 constellations were encountered.
166 The above burst-by-burst adaptive principles can also be extended to Adaptive Orthogonal Frequency
167 Division Multiplexing (AOFDM) schemes [17]. The associated AQAM principles were invoked in the lea context of parallel AOFDM modems also by Czylwik et al [18], Fischer [19] and Chow et al [20].
169 Our main contribution is that upon invoking the technique advocated - irrespective of the channel con-
,70 ditions experienced - the transceiver achieves always the best possible video quality by automatically
I?, adjusting the achievable bitrate and the associated video quality in order to match the channel quality ex-
,72 penenced This is achieved on a near-instantaneous basis under given propagation conditions in order to
,7 cater for the effects ot path-loss, tast-fading, slow-fading, dispersion, co-channel interference, etc Fur-
17 thermore, when the mobile is roaming in a hostile outdoor propagation environment, typically low-order
,75 low-rate modem modes are invoked, while in benign indoor environments predominantly the high-rate,
176 high source-signal representation quality modes are employed
I?? 3.2 ACDMA Signalling Scenarios i78 ACDMA transmission parameter adaptation is an action of the transmitter in response to time-varying
179 channel conditions It is only suitable for duplex communication between two stations, since the trans-
,80 mission parameter adaptation relies on some form of channel estimation and signalling In order to iβi efficiently react to the changes in channel quality, the following steps have to be taken
182 • Channel quality estimation In order to appropπately select the transmission parameters to be
183 employed for the next transmission, a reliable prediction of the channel quality duπng the next
184 active transmit timeslot is necessary
,85 • Choice of the appropriate parameters for the next transmission Based on the prediction of the las expected channel conditions duπng the next timeslot, the transmitter has to select the appropπate
187 modulation schemes for the subcarπers las • Signalling or blind detection of the employed parameters The receiver has to be informed, as
,89 to which set of demodulator parameters to employ for the received packet This information can
,90 either be conveyed within the packet, at the cost of loss of useful data bandwidth, or the receiver i9i can attempt to estimate the parameters employed at the transmitter by means of blind detection
192 mechanisms i93 Depending on the channel characteπstics, these operations can be performed at either of the duplex i94 stations, as shown in Figures 1(a), 1(b) and 1(c) If the channel is reciprocal, then the channel quality
195 estimation for each link can be extracted from the reverse link, and we refer to this regime as open-
196 loop adaptation In this case, the transmitter needs to communicate the transmission parameter set to i97 the receiver (Figure 1(a)), or the receiver can attempt blind detection of the transmission parameters 198 employed (Figure 1(c)) i99 If the channel is not reciprocal, then the channel quality estimation has to be performed at the receiver
2oo of the link In this case, the channel quality measure or the set of requested transmission parameters is
20, communicated to the transmitter in the reverse link (Figure 1(b)) This mode is referred to as closed-loop
202 adaptation
203 3.3 Video Transceiver
204 The schematic of the whole system is depicted m Figure 2 The multimedia source signal generated by
205 the video encoder of Figure 2 is assembled into transmission packets constituting a CDMA transmission 2oβ burst and the bits may be additionally mapped by the Mapper of Figure 2 to n number of different
207 Forward Error Correction (FEC) protection classes These bits are then convenveyed to the optional
208 Time Division Multiplex (TDMA)/ Time Division Duplex (TDD) scheme of Figure 2, before they are
209 assigned to the AQAM/ ACDMA modem seen in Figure 2
2io Again, the philosophy of the proposed burst-by-burst adaptive joint detection CDMA scheme is that the
2iι signal to interference plus noise ratio (SINR) at the output of the multi-user receiver is used in order to
2i2 estimate the instantaneous channel quality In one of its possible embodiments the receiver then decides
2i3 on the transmitter's mode to be used duπng the next transmission burst on the basis of the received signal
21 quality and the receiver's perception of the channel quality is signalled to the remote transmitter, in order
2i5 to allow it to satisfy the receiver's integnty requirement
2i6 In this study we transmitted 176x144 pixel Quarter Common Intermediate Format (QCEF) and 128x96
2i7 pixel Sub-QCIF (SQCIF) video sequences at 10 frames/s using a reconfigurable Time Division Multiple
2ia Access / Code Division Multiple Access (TDMA / CDMA) transceiver, which can be configured as a 1 ,
219 2 or 4 bit/symbol scheme shown in Figure 2 The H 263 video codec exhibits an impressive compression
220 ratio, although this is achieved at the cost of a high vulnerability to transmission errors, since a run-length
221 coded stream is rendered undecodable by a single bit error In order to mitigate this problem, when the __2 channel codec protecting the video stream is overwhelmed by the transmission errors, we refrain from
223 decoding the corrupted video packet in order to prevent error propagation through the reconstructed video
224 frame buffer [2] We found that it was more beneficial in video quality terms, if these corrupted video
225 packets were dropped and the reconstructed frame buffer was not updated, until the next video packet
226 replenishing the specific video frame area was received The associated video performance degradation
227 was found perceptually unobjectionable for packet dropping- or transmission frame error rates (FER)
228 below about 5% These packet dropping events were signalled to the remote decoder by supenmposmg
229 a strongly protected one-bit packet acknowledgement flag on the reverse-direction packet, as outlined
230 in [2] Bose-Chaudhuπ-Hocquenghem (BCH) and turbo error correction codes were used and again, 2 i the CDMA transceiver was capable of transmitting 1, 2 and 4 bits per symbol, where each symbol was 232 spread using a low spreading factor (SF) of 16, as seen in Table 1
Table 1 : Generic system parameters using the Frames spread speech/data mode 2 proposal [1 1]
-233 The associated parameters will be addressed in more depth during our further discourse. Employing
23 a low spreading factor of 16 allowed us to improve the system's multi-user performance with the aid
235 of joint-detection techniques [21]. We note furthermore that the implementation of the joint detection 23β receivers is independent of the number of bits per symbol associated with the modulation mode used,
23 since the receiver simply inverts the associated system matrix and invokes a decision concerning the
238 received symbol, irrespective of how many bits per symbol were used. Therefore, joint detection
239 receivers are amenable to amalgamation with the above 1, 2 and 4 bit/symbol modem, since they
240 do not have to be reconfigured each time the modulation mode is switched.
24i In this performance study we used the Pan-European FRAMES proposal [11] as the basis for our CDMA
2 2 system. The associated transmission frame structure is shown in Figure 3, while a range of generic system
Table 2: FEC-protected and unprotected BCH and Turbo coded bitrates for the 4QAM transceiver mode
parameters are summarised in Table 1. In our performance studies we used the COST207 seven-path bad urban (BU) channel model, whose impulse response is portrayed in Figure 4.
Our initial experiments compared the performance of a whitening matched filter (WMF) for single user detection and ώe Minimum mean square error block decision feedback equalizer (MMSE-BDFE) for joint multi-user detection. These simulations were performed using 4-level Quadrature Amplitude Modulation (4QAM), transmitting both binary BCH and turbo coded video packets. The associated bitrates are summarised in Table 2.
The transmission bitrate of the 4QAM modem mode was 29.5Kbps, which was reduced due to the approximately half-rate BCH or turbo coding, plus the associated video packet acknowledgement feedback flag error control and video packetisation overhead to produce effective video bitrates of 13.7Kbps and 1 1.1Kbps, respectively. A more detailed discussion on the video packet acknowledgement feedback error control and video packetisation overhead will be provided in Section 3.4 with reference to the convolutionally coded multi-mode investigations.
Figure 5 portrays the bit error ratio (BER) performance of the BCH coded video transceiver using both matched filtering and joint detection for 2-8 users. The bit error ratio is shown to increase, as the number of users increases, even upon employing the MMSE-BDFE multi-user detector (MUD). However, while the matched filtering receiver exhibits an unacceptably high BER for supporting perceptually unimpaired video communications, the MUD exhibits a far superior BER performance.
When the BCH codec was replaced by the turbo-codec, the bit error ratio performance of both matched filtering and the MUD receiver improved, as shown in Figure 6. However, as expected, matched filtering was still outperformed by the joint detection scheme for the same number of users. Furthermore, the matched filtering performance degraded rapidly for more than two users.
Figure 7 shows the video packet loss ratio (PLR) for the turbo coded video stream using matched filtering and joint detection for 2-8 users. The figure clearly shows that the matched filter was only capable of meeting the target packet loss ratio of 5% for upto four users, when the channel SNR was in excess of 1 ldB. However, the joint detection algorithm guaranteed the required video packet loss ratio performance
Table 3: Operational-mode specific transceiver parameters for the proposed multi-mode system
26 for 2-8 users in the entire range of channel SNRs shown. Furthermore, the 2-user matched-filtered PLR
27 performance was close to the 8-user MUD PLR.
27i 3.4 Multi-mode Video System Performance
-72 Having shown that joint detection can substantially improve our system's performance, we investigated j73 the performance of a multi-mode convolutionally coded video system employing joint detection, while
274 supporting two users. The associated convolutional codec parameters are summarised in Table 3.
275 Below we now detail the video packetisation method employed. The reader is reminded that the number
276 of symbols per TDMA frame was 68 according to Table 1. In the 4QAM mode this would give 136 bits
277 per TDMA frame. However, if we transmitted one video packet per TDMA frame, then the packetisation
278 overhead would absorb a large percentage of the available bitrate. Hence we assembled larger video
279 packets, thereby reducing the packetisation overhead and arranged for transmitting the contents of a 2βo video packet over three consecutive TDMA frames, as indicated in Table 1. Therefore each protected 28i video packet consists of 68 x 3 = 204 modulation symbols, yielding a transmission bitrate of between 282 14.7 and 38.9 Kbps for BPSK and 16QAM, respectively. However, in order to protect the video data
-3 we employed halt-rate, constraint-length nine convolutional coding, using octal generator polynomials
-84 ot 561 and 753 The useful video bitrate was further reduced due to the 16-bit Cyclic Redundancy ess Checking (CRC) used tor error detection and the nine-bit repetition-coded feedback error flag for the
286 reverse link This results in video packet sizes ot 77, 179 and 383 bits for each ot the three modulation
287 modes The useful video capacity was finally further reduced by the video packet header of between 8
288 and 10 bits, resulting in useful or effective video bitrates ranging from 5 to 26 9 Kbps in the BPSK and
289 16QAM modes, respectively
290 The proposed multi-mode system can switch amongst the 1, 2 and 4 bit/symbol modulation schemes 29, under network control, based upon the prevailing channel conditions As seen in Table 3, when the
292 channel is benign, the unprotected video bitrate will be approximately 26 9Kbps the 16QAM mode
293 However, as the channel quality degrades, the modem will switch to the BPSK mode ot operation, where j4 the video bitrate drops to 5Kbps, and for maintaining a reasonable video quality, the video resolution has
295 to be reduced to SQCIF (128x96 pels)
296 Figure 8 portrays the packet loss ratio for the multi-mode system, in each ot its modulation modes tor a
297 range ot channel SNRs It can be seen in the figure that above a channel SNR of !4dB the 16QAM mode
298 offers an acceptable packet loss ratio of less than 5%, while providing an unprotected video rate of about
299 26 9Kbps If the channel SNR drops below 14dB, the multi-mode system is switched to 4QAM and 00 eventually to BPSK, when the channel SNR is below 9dB, in order to maintain the required quality of 3oι service, which is dictated by the packet loss ratio The figure also shows the acknowledgement feedback
302 error ratio (FBER) for a range of channel SNRs, which has to be substantially lower, than the video 0 PLR itself This requirement is satisfied in the figure, since the feedback eπors only occur at extremely
304 low channel SNRs, where the packet loss ratio is approximately 50%, and it is therefore assumed that s the multi-mode system would have switched to a more robust modulation mode, before the feedback
3oβ acknowledgement flag can become corrupted so? The video quality is commonly measured in terms of ώe peak-signal-to-noise-ratio (PSNR) Figure 9
308 shows the video quality in terms of the PSNR versus ώe channel SNRs for each of the modulation
309 modes As expected, the higher throughput bitrate of the 16QAM mode provides a better video quality 3io However, as ώe channel quality degrades, the video quality of the 16QAM mode is reduced and hence 3iι it becomes beneficial to switch from the 16QAM mode to 4QAM at an SNR of about 14dB, as it was i2 suggested by the packet loss ratio performance of Figure 8 Although the video quality expressed in 3i3 terms ot PSNR is supeπor tor the 16QAM mode in compaπson to the 4QAM mode at channel SNRs 3i in excess of 12dB, however, due to the excessive PLR the perceived video quality appears infeπor in sis compaπson to that of the 4QAM mode, even though the 16QAM PSNR is higher for channel SNRs
i6 in the range ot 12-14dB More specifically, we found that it was beneficial to switch to a more robust si? modulation scheme, when the PSNR was reduced by about ldB with respect to its unimpaired PSNR
3i8 value This ensured that the packet losses did not become subjectively apparent, resulting in a higher sis perceived video quality and smoother degradation, as the channel quality deteπorated
320 The effect ot packet losses on the video quality quantified in terms of PSNR is portrayed in Figure 10
321 The figure shows, how the video quality degrades, as the PLR increases It has been found that in order
322 to ensure a seamless degradation of video quality as the channel SNR reduced, it was the best policy
323 to switch to a more robust modulation scheme, when the PLR exceeded 5% The figure clearly shows 2 that a 5% packet loss ratio results in a loss of PSNR, when switching to a more robust modulation
325 scheme However, if the system did not switch until the PSNR of the more robust modulation mode
326 was similar, the perceived video quality associated with the oπginally higher rate, but channel-impaired 2r stream became infeπor
32β 3.5 Burst-by-Burst adaptive videophone system 29 A burst-by-burst adaptive modem, maximizes the system's throughtput by using the most appropπate
330 modulation mode for the cuπent instantaneous channel conditions Figure 1 1 exemplifies, how a burst- 3i by-burst adaptive modem changes its modulation modes based on the fluctuating channel conditions The 32 adaptive modem uses the SINR estimate at ώe output of the jo t-detector to estimate the instantaneous 33 channel quality, and hence to set the modulation mode 34 The probability of the adaptive modem using each modulation mode for a particular channel SNRs is
335 portrayed in Figure 12 It can be seen at high channel SNRs that the modem mainly uses the 16QAM
336 modulation mode, while at low channel SNRs the BPSK mode is most prevalent
_37 The advantage of dynamically reconfigured burst-by-adaptive modem over the statically switched multi-
33β mode system previously descπbed, is that the video quality is smoothly degraded as the channel condi-
339 tions detenorate The switched multi-mode system results in more sudden reductions in video quality, 0 when the modem switches to a more robust modulation mode Figure 13 shows the throughput bitrate 4i of the dynamically reconfigured burst-by-burst adaptive modem, compared to the three modes of the 42 statically switched multi-mode system The reduction of the fixed modem modes' effective throughput 43 at low SNRs is due to the fact that under such channel conditions an increased fraction of the transmitted
3 4 packets have to be dropped, reducing the effective throughtput The figure shows the smooth reduction of 45 the throughput bitrate, as the channel quality deteπorates The burst-by-burst modem matches the BPSK 34β mode's bitrate at low channel SNRs, and the 16QAM mode's bitrate at high SNRs The dynamically 3 reconfigured burst-by-burst adaptive modem characteπsed in the figure perfectly estimates the prevalent
1 8 channel conditions although in practice the estimate ot channel quality is not perfect and it is inherentiv j j delayed However, we have found that non-pertect channel estimates result in only slightly reduced
350 performance, when compared to perfect channel estimation
35i The smoothly varying throughput bitrate of the burst-by-burst adaptive modem translates into α smoothly 52 varying video quality as the channel conditions change The video quality measured in terms of the
35 average peak signal to noise ratio (PSNR) is shown versus ώe channel SNR in Figure 14 in contrast to
354 that of the individual modem modes The figure demonstrates that the burst-by-burst adaptive modem ass provides equal or better video quality over a large proportion of the SNR range shown than the individual
356 modes However, even at channel SNRs, where the adaptive modem has a slightly reduced PSNR, the
357 perceived video quality ot the adaptive modem is better since the video packet loss rate is far lower, than 5β that ot the fixed modem modes
9 Figure 15 shows the video packet loss ratio versus channel SNR for the three fixed modulation modes
360 and the burst-by-burst adaptive modem with perfect channel estimation Again the figure demonstrates 6i that the video packet loss ratio of the adaptive modem is similar to that of the fixed BPSK modem mode,
362 however the adaptive modem has a far higher bitrate throughput, as the channel SNR increases The 63 burst-by-burst adaptive modem gives an error performance similar to that of the BPSK mode, but with
364 the flexibtty to increase the bitrate throughput of the modem, when the channel conditions improve If
365 imperfect channel estimation is used, the throughput bitrate of the adaptive modem is reduced slightly 66 Furthermore, the video packet loss ratio seen in Figure 15 is slightly higher for the AQAM scheme due
367 to invoking higher-order modem modes, as the channel quality increases However we have found that 3β8 is possible to maintain the video packet loss ratio within tolerable limits for the range of channel SNRs 369 considered o The interaction between the video quality measured m terms of PSNR and the video packet loss ratio
37i can be more clearly seen in Figure 16 The figure shows that ώe adaptive modem slowly degrades
372 ώe decoded video quality from that of the error free 16QAM fixed modulation mode, as ώe channel
373 conditions detenorate The video quality degrades from the error-free 41dB PSNR, while maintaining a
374 near-zero video packet loss ratio, until the PSNR drops below about 36dB PSNR At this point the further
375 reduced channel quality inflicts an increased video packet loss rate and the video quality degrades more
376 slowly The PSNR versus packet loss ratio performance then tends toward that achieved by the fixed
377 BPSK modulation mode However the adaptive modem achieved better video quality than the fixed 37a BPSK modem even at high packet loss rates
379 3.6 Summary
380 A joint-detection assisted multimode CDMA-based video transceiver was proposed, which substantially 3βι outperformed the conventional matched-filteπng based transceiver, which was characteπsed by adap- 382 tively reconfiguπng the transceiver's mode of operation based on the instantaneous channel quality In
333 our transceiver a higher number of bits per modulation symbol was invoked by the transmitter, when
33 the channel quality was sufficiently high for supporting this more bitrate efficient, but less eπor resilient
385 transmission mode By contrast, a more error resilient but less bitrate efficient mode was invoked for
386 supporting error-free CDMA transmission over wireless multi-user channels
387 In this embodiment the above property was exploited in a practical adaptive video transceiver, which 88 instructed the video codec to generate the required number of bits that the CDMA transceiver was capable
389 ot deliveπng in its current channel-condition dependent configuration mode
390 In other embodiments the proposed burst-by-burst adaptive transceiver can be invoked in the context 39i of arbitrary multimedia signals, irrespective of their resolution or source representation quality Spe-
392 cific further embodiments of such codecs are constituted by programmable-rate speech, audio, video,
393 handwπting codecs, which can be configured by the transceiver to generate a channel-quality dependent
394 number of source-coded bits
395 The proposed burst-by-burst adaptive video transceiver guaranteed a near-unimpaired video quality for 39β channel SNRs in excess of about 5 dB over the COST207 dispersive Rayleigh-faded channel The ben- 397 efits of the multimode video transceiver clearly manifest themselves in terms of supporting un-impaired 39β video quality under time-vaπant channel conditions, where a single-mode transceiver's quality would
399 become severely degraded by channel effects The dynamically reconfigured burst-by-burst adaptive
400 modem gave better perceived video quality due to its more graceful reduction in video quality, as the -oι channel conditions degraded, than a statically switched multi-mode system
402 References
403 [1] J Streit and L Hanzo, "Dual-mode vector-quantised low-rate cordless videophone systems for
404 indoors and outdoors applications," IEEE Tr on Vehicular Technology, vol 46, pp 340-357, May
405 1997 06 [2] P Cherπman and L Hanzo, "Programmable H 263-based wireless video transceivers tor
407 interference-limited environments," IEEE Trans on Circuits and Systems for Video Technology,
408 vol 8, pp 275-286, June 1998
09 [3] N. Fαrber, E Steinbach, and B Girod, "Robust h.263 compatible transmission for mobile video
4io server access," in Proc of First International Workshop on Wireless Image/Video Communications. iι (Loughborough, UK), pp 122-124, 4-5 September 1996.
4i2 [4] A. Sadka, F Eryurtlu, and A. Kondoz, "Improved performance H.263 under erroneous transmission
4,3 conditions," Electronics Letters, vol. 33, pp. 122-124, Jan 16 1997
4i4 [5] W Webb and R. Steele, "Vaπable rate QAM for mobile radio," IEEE Transactions on Commum-
4,5 cations, vol. 43, no. 7, pp. 2223-2230, 1995.
4i6 [6] H Matsuoka, S. Sampei, N Moπnaga, and Y. Kamio, "Adaptive modulation system with vaπable
4i7 coding rate concatenated code for high quality multi-media communications systems," in Proceed-
4,β ings of IEEE VTC '96, (Atlanta, GA, USA), pp. 487-491, IEEE, 1996.
4i9 [7] J Torrance and L. Hanzo, "Upper bound performance of adaptive modulation in a slow Rayleigh
420 fading channel," Electronics Letters, vol. 32, pp. 718-719, 11 Apπl 1996.
42i [8] J. Torrance and L. Hanzo, "Optimisation of switching levels for adaptive modulation in a slow
422 Rayleigh fading channel," Electronics Letters, vol. 32, pp. 1 167-1 169, 20 June 1996
423 [9] J Torrance and L. Hanzo, "Demodulation level selection in adaptive modulation," Electronics Let-
424 ters, vol 32, pp. 1751-1752, 12 September 1996.
425 [10] A. Goldsmith and S. Chua, " Vaπable Rate Vaπable Power MQAM for Fading Channels," IEEE
426 Transactions on Communications, vol. 45, pp. 1218 - 1230, October 1997 27 [11] A. Klein, R. Pirhonen, J. Skoeld, and R. Suoranta, "FRAMES multiple access mode I - wideband .a TDMA with and wiώout spreading," in Proceedings of IEEE International Symposium on Personal,
429 Indoor and Mobile Radio Communications, PIMRC'97, vol. 1, (Manna Congress Centre, Helsinki,
430 Finland), pp. 37-41, IEEE, 1-4 Sept 1997.
43i [12] J Torrance and L. Hanzo, "Latency and networking aspects of adaptive modems over slow indoors
432 rayleigh fading channels," IEEE Jr. on Veh. Techn., vol. 48, no. 4, pp. 1237-1251, 1998
43 [13] J. Torrance, L. Hanzo, and T. Keller, "Interference aspects of adaptive modems over slow rayleigh
434 fading channels," IEEE Tr on Veh. Techn., vol. 48, pp. 1527-1545, Sept 1999
435 [14] V. Lau and M. Macleod, "Vanable rate adaptive trellis coded QAM for high bandwidth efficiency
436 applications in rayleigh fading channels," in Proceedings of IEEE Vehicular Technology Conference
437 (VTC'98), (Ottawa, Canada), pp. 348-352, IEEE, May 1998
38 [ 1 ] C. Wong, T Liew, and L. Hanzo, "Blind modem mode detection aided block turbo coded burst-by-
439 burst wideband adaptive modulation," in Proceeding of ACTS Mobile Communication Summit '99.
440 (Sorrento, Italy), ACTS, June 8-1 1 1999.
4, [ 16] M. Yee and L. Hanzo, "Upper-bound performance of radial basis function decision feedback
442 equalised burst-by-burst adaptive modulation," in Proceedings of ECMCS'99, (Krakow, Poland),
443 24-26 June 1999.
44 [17] T. Keller and L. Hanzo, "Adaptive orthogonal frequency division multiplexing schemes," in Pro-
445 ceeding of ACTS Mobile Communication Summit '98 [22], pp. 794-799.
46 [18] A. Czylwik, "Adaptive OFDM for wideband radio channels," in Proceeding of IEEE Global •47 Telecommunications Conference, Globecom 96 [23], pp. 713-718. 4β [19] R. Fischer and J. Huber, "A new loading algorithm for discrete multitone transmission," in Pro-
449 ceeding of IEEE Global Telecommunications Conference, Globecom 96 [23], pp. 713-718.
450 [20] P. Chow, J. Cioffi, and J. Bingham, "A practical discrete multitone transceiver loading algorithm 45i for data transmission over spectrally shaped channels," IEEE Trans, on Communications, vol. 48,
452 pp. 772-775, 1995.
453 [21] E. Kuan and L. Hanzo, "Joint detection CDMA techniques for third-generation transceivers," in
454 Proceeding of ACTS Mobile Communication Summit '98 [22], pp. 727-732.
455 [22] ACTS, Proceeding of ACTS Mobile Communication Summit '98, (Rhodes, Greece), 8-11 June
456 1998.
457 [23] IEEE, Proceeding of IEEE Global Telecommunications Conference, Globecom 96, (London, UK),
458 18-22 Nov 1996.