Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025 www.elsevier.de/aeue

Integrated voice and data transmission employing adaptive modulation in wireless networks Rajarshi Mahapatra∗ , Anindya Sundar Dhar, Debasish Datta Department of Electronics and Electrical Communication Engineering, Indian Institute of Technology, Kharagpur 721 302, India Received 7 September 2007; accepted 22 August 2008

Abstract In this paper, we propose an integrated voice and data transmission technique using adaptive modulation, wherein the effective data transmission rate (throughput) is improved by allowing the bit-error rate (BER) for voice to increase beyond the BER needed for data. In a given integrated frame with data and voice bits, voice bits are allowed to use higher-order modulation as compared to data bits due to the higher tolerance of voice to BER. This leads to a larger symbol duration for voice which is subsequently shrunk into fewer bits, provided the resulting increase in voice BER does not exceed its maximum acceptable limit. This shrinking operation releases some bits from the voice quota, which are eventually utilized for data transmission in addition to the predefined data bits in the same frame. In this work, we first examine the feasibility of the proposed scheme for a time-invariant AWGN channel. Subsequently, we use our methodology in an adaptive manner for fading channels. We also examine the performance of the proposed scheme for a Global System for Mobile communication (GSM) system. Using the proposed scheme, our analysis indicates that one can dynamically enhance the overall throughput significantly in a broadband wireless network. 䉷 2008 Elsevier GmbH. All rights reserved. Keywords: Adaptive modulation; Integrated voice and data systems

1. Introduction Continued increase in demand for all types of wireless services (voice, data, and multimedia) has been fueling the need for a higher capacity and a lower bit-error rate (BER) in wireless networks. However, the wireless networks lag behind the wireline communication networks in terms of transmission quality. To fulfill the evolving demands in wireless networks with transmission challenges (e.g., limited bandwidth, fading, power efficiency for mobile units, etc.) [1,2], several techniques have been proposed from time to time. By utilizing the available information on the timevarying nature of wireless channels, a variety of adaptive ∗ Corresponding author.

E-mail address: [email protected] (R. Mahapatra). 1434-8411/$ - see front matter 䉷 2008 Elsevier GmbH. All rights reserved. doi:10.1016/j.aeue.2008.08.002

transmission techniques have been developed to improve the spectral efficiency of wireless systems. These techniques assume that the channel quality is known at both transmitter and receiver ends. According to the channel quality, these adaptive schemes change appropriate transmission parameters, such as transmitted power, symbol rate, constellation size, coding rate, or any combination of these parameters [3–13]. In general, voice transmission has low data rate requirements with real-time delay constraints, while data transmission demands higher rates with less stringent delay requirements, but video transmission needs a higher data rate along with real-time delay constraints. This suggests that low-fixed-rate transmission combined with power adaptation, where the transmitter adjusts its power to maintain a constant carrier-to-noise ratio (CNR) at the receiver, is

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

well suited to voice, while bursty variable-rate transmission, which maximizes average spectral efficiency, is best suited to data communication. On the other hand, high data rate transmission combined with power adaptation is well suited for video. In addition, voice, data, and video typically have different BER requirements, which must be incorporated into their respective transmission schemes. According to European Telecommunications Standards Institute (ETSI) specification, General Packet Radio Service (GPRS) and Enhanced Data Rates for Global System for Mobile communication (GSM) Evolution (EDGE) support the simultaneous voice and data communication (SVDC) for class-A and class-B users [14]. In a GPRS network, the mobile user enjoys SVDC by the dynamic allocation of the TDMA slots. In GSM/GPRS, a maximum of four TDMA slots are allocated in uplink and downlink in order to increase the data rate. Based on different multislot classes, these four TDMA slots in both ways are dynamically allocated among the mobile users. These multislot classes are product dependent, and determine the maximum achievable data rates in uplink and downlink directions. Furthermore, to support a high bit rate, future wireless communication systems can provide SVDC along with adaptive modulation and coding [15–19]. One such technique was proposed in [20], where a significant improvement in performance was obtained by making use of speech modeling either at the receiver or at the transmitter. When it is used at the receiver, such modeling aids the data decoding process. In [21], Alouini et al. proposed an SVDC scheme to take advantage of the time-varying nature of fading to dynamically allocate the transmitted power between the inphase and quadrature channels. In this paper, they used fixedrate binary phase shift keying (BPSK) modulation on the quadrature channel for voice, and variable-rate M-ary amplitude modulation (M-AM) on the inphase channel for data. In order to improve the spectral efficiency, an adaptive technique employing a variable rate uniform M-ary quadrature amplitude modulation (M-QAM) was proposed in [22] for simultaneous voice and a single-class data transmission. In this work, Hwang et al. applied different switching thresholds according to the required BER. Specifically, it starts to transmit voice with BPSK when the CNR allows BPSK transmission with the target BER for voice. When the CNR is improved to support M-QAM with the target BER for data, it begins to transmit both data and voice with the same BER target. In [23], Hossain et al. proposed a technique using adaptive hierarchical modulation for simultaneous voice and multi-class data transmission over fading channels. They also extended the work proposed in [21] to make the system spectrally efficient. It may be noted that in order to support adaptive modulation, a wireless system needs the feedback information from the receiver about the channel condition, which is estimated by channel estimation techniques. However, delay in estimation, channel estimation error, and error in feedback information might impair the adaptive scheme, which we have not considered in our work [24–26].

1013

In this paper, we propose an integrated (i.e., simultaneous) voice and data transmission technique using an adaptive modulation scheme, wherein one can employ a dynamic capacity allocation for voice and data bits to improve overall spectral efficiency (and hence network throughput). In the proposed scheme the effective transmission rate for data bits is enhanced by allowing the BER for voice to increase beyond the BER needed for data. In a given integrated frame having data and voice bits, voice bits are allowed to use higher-order modulation as compared to data bits due to the higher tolerance of voice to BER. This leads to a larger symbol duration for voice which is subsequently compressed (shrunk) into fewer bit intervals, provided the resulting increase in voice BER does not exceed its maximum acceptable limit. This shrinking operation releases some bit intervals from the voice quota, which are eventually utilized for the transmission of data bits in addition to the preallocated data bits in the same frame. With this technique, the transmitter is able to send more data bits in a given time frame, while maintaining the same number of voice bits (albeit with an acceptable increase in voice BER due to the higher-order modulation and symbol compression). However, this operation is carried out if and only if the BER of the transmission system is suitable for data transmission. The rest of this paper is organized as follows. Section 2 presents the proposed scheme to improve the data transmission rate in a BER-aware integrated voice and data communication technique. The practical implementation of the proposed scheme is also described in this section. In Section 3, we present the numerical results on the performance of the proposed scheme for a time-invariant AWGN channel. For a fading channel, we use our methodology in an adaptive manner and find the performance of the proposed scheme in Section 4. In Section 5, the performance of the proposed scheme is examined in a GSM-TDMA system. Finally, Section 6 presents summary of our work on this topic.

2. Capacity allocation for voice and data using BER-aware adaptive modulation In this section, we describe an integrated voice and data transmission scheme for improving the data throughput. In the proposed scheme, we consider a time frame that contains a certain number of bits, each with a bit interval of T, wherein some bit intervals are used for voice communication and the rest for data transmission. For successful transmission, voice and data have different upper limits of BER and, voice can be successfully transmitted with higher BER as compared to data. Therefore, in a given integrated frame with data and voice bits, voice bits can be allowed to use higher-order modulation as compared to data bits and the symbol intervals for the chosen higher-order modulation can also be shrunk unless the voice BER does not exceed its maximum acceptable limit (BERmax (voice)). Thus

1014

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

100 a b c d e f

10−1 10−2

BER

10−3 10−4 10−5 10−6 10−7 10−8

10

15

20

25 CNR (dB)

30

35

40

Fig. 1. BER performance of 64-QAM modulation with different values of symbol duration (Ts ). a: Ts = 6T , b: Ts = 5T , c: Ts = 4T , d: Ts = 3T , e: Ts = 2T , f: Ts = T .

our aim is to employ higher-order modulation for voice and shrink the voice-symbol interval to some extent by using adaptive modulation and allow the voice BER to increase, which does not exceed BERmax (voice). Having carried out this task, one can eventually save some bit intervals from the voice field and allocate these additional bits for data transmission. Therefore, we propose to use a higher constellation size for voice and then shrink the symbol duration for voice by a few bit intervals, thereby gaining in spectral efficiency and data throughput by using the additional bits of the voice field for data transmission in a given integrated frame. To explain the proposed scheme in detail, we consider a communication link with a BER BERmax (data), where BERmax (data) is the maximum acceptable BER for successful data transmission (we assume that BERmax (voice) is always greater than BERmax (data)). Hence, the voice and data can be communicated over this channel between two end nodes, since the voice can tolerate a higher BER than the data. Thus, at a particular BER  BERmax (data), system can employ in general a suitable M-ary QAM and select the same constellation size for voice and data transmission, i.e., Mv = Md , where Mv and Md are the constellation sizes for voice and data, respectively. As mentioned earlier, by exploiting the nature of higher tolerance to BER, voice bits can now choose the higher constellation size using adaptive modulation while ensuring that the voice BER does not exceed BERmax (voice). This would make ?he constellation size of voice larger than the data constellation size, i.e., Mv > Md . With this consideration, voice bits can be transmitted using the constellation size Mv of symbol interval Ts (=kv T, kv = log2 Mv ). Next the shrinking of symbol

interval of voice constellation can take place by reducing the symbol interval from Ts to Ts−sh = Ts − T, Ts − 2T, . . . etc.. This operation becomes feasible if the impact of increasing the constellation size followed by shrinking operation does not push the voice BER above the acceptable limit BERmax (voice). Following this operation, the transmitted voice symbols take lesser transmission time (as compared to the unshrunk symbol) to transmit the same information, which in turn releases some bit intervals from the voice segment of the frame, which are eventually used to transmit some additional data bits. It is important to note that the proposed scheme will work if the receiver BER is below its maximum acceptable limit BERmax (data), which may be typically 10−7 (say) [27,28]. Having ensured this with an appropriate Md (for a given CNR), one can choose a value of Mv  Md and shrink the voice-symbol intervals from Ts to Ts−sh < Ts . It is therefore worthwhile to examine how the receiver BER would increase (for voice bits) if the symbol duration is shrunk from Ts to Ts−sh for a given Mv . Fig. 1 shows the BER performance of 64-QAM modulation with different symbol durations. It is found that as the symbol duration of 64-QAM shrinks from 6T to T, BER increases steadily for a given CNR (see the Appendix for details). Using these results, while shrinking the symbol duration, one therefore needs to ensure that the degraded BER does not exceed BERmax (voice). It may be noted that in this work, we considered the system without channel coding. The implementation of the proposed scheme is described in the following in further details. In the first step, we calculate the suitable constellation sizes for voice (Mv )

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

and data (Md ) for specified values of BERmax (voice) and BERmax (data) at the prevailing receiver CNR using Eq. (8) (see the Appendix). In Eq. (8), we consider Mi = Mv or Md , where i represents v and d for voice and data respectively, BER represents BERmax (voice) or BERmax (data) of an M-QAM system in an AWGN channel, and  is the received CNR. It may be noted that the value of Mi may not be equal to a feasible constellation size. To make it a feasible number, we rounddown Mi to Mˆ i as Mˆ i = 2log2 Mi  and ki = log2 Mˆ i 

(1)

where ki (=kv or kd ) is the number of bits per symbol (voice or data) of the corresponding constellation size. To shrink the symbol duration, we first calculate the BER for the modified values, Mˆ v and Mˆ d , which might decrease below Mv and Md due to a rounddown operation. If Mˆ d  the lowest permissible constellation size Mdo (say) and the corresponding BER for data  BERmax (data), the shrinking operation is undertaken for voice transmission. Otherwise, the shrinking operation as well as the data transmission becomes infeasible and voice transmission takes place with these appropriate constellation size and symbol duration. It may be noted that the shrinking operation for voice symbols with a given Mˆ v as the constellation size can be carried out, provided that the corresponding BER for voice is below BERmax (voice). However, the BER value corresponding to Mˆ v may be very close to the BERmax (voice), and hence an attempt to shrink the corresponding symbol interval Ts (=kv T ) to Ts−sh (=kv−sh T , kv−sh < kv ) might push the voice BER beyond BERmax (voice). In view of this, having estimated Mˆ v in the first step along with a search for a possible Ts−sh , it would often be more useful to explore next with the lower possible value for constellation size Mv = Mˆ v /2 (kv = log2 Mv , kv kv ) and then attempt the

1015

shrinking operation on symbol duration kv T with kv reduced  to kv−sh < kv . With a lower value of constellation size, one is expected to gain some margin in BER, which can be subsequently utilized for shrinking the symbol duration in steps of T The shrinking of symbol interval Ts is continued until the BER remains below BERmax (voice), and thus eventually  we get the minimum symbol interval Ts−sh for a particular  constellation size Mv . However, it may be noted that Mv should not be reduced below 4 (corresponding to a symbol duration=2T ), because the next lower value of Mv (=2, corresponding to a symbol duration = T ) will not permit any shrinking operation. Thus, to make the symbol shrinking feasible, we would in general set up a lower limit Mvo for Mv , which would be in general 4 (corresponding to QPSK for the m-ary PSK family). However, for data transmission, since we do not propose to carry out any shrinking, the minimum value for constellation size can go down to 2 (i.e., Mvo = 2, corresponding to BPSK for the m-ary PSK family) if necessary.  With kv−sh (min), representing the minimum value of  Ts−sh at a given constellation size Mv , we define the shrinking factor (SF) for voice transmission as SF =

 kv−sh (min)

(2)

log2 (Mv )

For each constellation size Mv from Mˆ v to 4 ( Mˆ v , Mˆ v /2, Mˆ v /4 . . ., with 4 for QPSK), we find out the min (min) imum value for the shrunk symbol interval kv−sh and the corresponding SF. Thereafter, we finally select a  specific combination of Mv and the corresponding kv−sh   (hereafter designated as Mv (final), kv−sh (final)), that offers the minimum value for SF (i.e., maximum shrinking). If the integrated (voice and data) frame consists of N (=Nv + Nd ) bits with Nv bits for voice and

Table 1. Transmission strategy of integrated voice and data frame Received CNR (dB)

Mˆ v

kv

Mˆ d

kd

Transmission status

Feasibility of shrinking

5

0

0

0

0

No transmission

Not applicable

10 12 14

4 4 8

2 2 3

0 0 0

0 0 0

Voice only BERmax (data) < BER  BERmax (voice)

Shrinking not allowed

15 17 20 22 25 28 30 32 35 40

8 16 16 32 64 128 256 256 256 256

3 4 4 5 6 7 8 8 8 8

4 4 8 16 32 64 64 128 256 256

2 2 3 3 4 5 5 6 7 8

Voice and data BER  BERmax (data)

Allowed

1016

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

Nd bits for data, the percentage increase in data throughput through this scheme can be expressed as  =

(1 − SF)Nv × 100% Nv + Nd

(3)

Using these expressions for SF and , we evaluate the performance of the proposed scheme in the following.

3. Performance in a time-invariant AWGN channel In this section, the feasibility of the proposed technique is first examined for a time-invariant AWGN channel (i.e., for

a static case without fading) using numerical results. Subsequently, in the next section, we extend the investigation for a fading channel with a provision for adaptive modulation. We consider the typical values for BERmax (voice) and BERmax (data) as 10−3 and 10−7 , respectively, and the minimum realizable constellation size as 4. We also assume that the integrated frame contains an equal number of bits for voice and data (i.e., Nv = Nd = N /2), as a specific case. Table 1 shows the constellation size of voice and data transmission for different values of received CNR. As evident from Table 1, based on BER calculation, only the voice communication is supported when CNR  10 dB. However, the system enables data transmission along with voice, when CNR  15 dB.

Table 2. Selection of a reduced symbol interval on possible voice constellation size for a CNR of 17 dB Received CNR (dB)

Mˆ v

Mv

kv

Search for  (min) kv−sh

BER after shrinking

SF

17

16

16

4

4 3 2 1

5.7e−4 2.2e−3 9.4e−3 4.2e−2

1.00 0.75 0.50 0.25

8

3

3 2 1

1.5e−6 6.6e−5 3.2e−3

1.00 0.67 0.33

4

2

2 1

7.2e−13 2.7e−7

1.00 0.50

Table 3. Selection of a reduced symbol interval on possible voice constellation size for a CNR of 25 dB Received CNR (dB)

Mˆ v

Mv

kv

Search for  (min) kv−sh

BER after shrinking

SF

25

64

64

6

6 5 4 3 2 1

3.0e−5 1.1e−4 4.4e−4 1.7e−3 7.3e−3 3.3e−2

1.00 0.83 0.67 0.50 0.33 0.17

32

5

5 4 3 2 1

1.0e−8 2.4e−7 6.0e−6 1.5e−4 4.4e−3

1.00 0.80 0.60 0.40 0.20

16

4

4 3 2 1

6.8e−16 2.1e−12 7.0e−9 2.6e−5

1.00 0.75 0.50 0.25

8

3

3 2 1

1.0e−31 8.5e−22 7.7e−12

1.00 0.67 0.33

4

2

2 1

4.8e−71 1.4e−36

1.00 0.50

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

1017

 Table 4. Results of the proposed scheme for different values of CNR (Mv and kv−sh in bold face indicate the values of Mv (final) and  kv−sh (final) offering minimum values for SF)

kv

 kv−sh

8 4

3 2

4

16 8 4

16

8

22

32

25

Mˆ v

Mˆ d

Mv

15

8

4

17

16

20

Received CNR (dB)

BER after shrinking

SF

3 1

9.9e−5 3.4e−5

1.00 0.50

4 3 2

4 2 1

5.7e−4 6.6e−5 2.7e−7

1.00 0.67 0.50

16 8 4

4 3 2

2 1 1

5.8e−4 6.7e−5 7.6e−13

0.50 0.33 0.50

16

32 16 8

5 4 3

3 2 1

7.9e−4 2.5e−5 8.4e−7

0.60 0.50 0.33

64

32

64 32 16

6 5 4

4 2 1

4.4e−4 1.5e−4 2.6e−5

0.67 0.40 0.25

28

128

64

128 64 32

7 6 5

4 2 1

9.1e−4 4.5e−4 1.5e−4

0.57 0.33 0.20

30

256

64

256 128 64 32

8 7 6 5

6 3 2 1

6.9e−4 3.8e−4 1.9e−5 3.5e−6

0.75 0.42 0.33 0.20

32

256

128

256 128 64

8 7 6

4 2 1

5.3e−4 2.7e−4 1.1e−4

0.50 0.28 0.16

35

256

256

256 128

8 7

2 1

5.3e−4 2.8e−4

0.25 0.14

40

256

256

256

8

1

2.9e−5

0.12

The recursive process of the proposed scheme for choos ing the final values for Mv (final) and kv−sh (final) is described in Table 2 for a typical case. As evident from Table 2, at a given CNR of 17 dB, the voice bits can support 16-QAM modulation for successful communication with a BER of 5.7 × 10−4 . Hence, the shrinking operation is explored from 16-QAM modulation down to 4-QAM and for each constellation size, we shrink the symbol interval down to a single bit interval. At each step, the BER and SF are evaluated. After going through the entire exercise, we select the final constellation size Mv (final) and reduced symbol  interval kv−sh (final) corresponding to the lowest SF. In Table 2, we observe that the lowest SF is 0.25, but the corresponding BER > 10−3 , which is not suitable for voice communication. Therefore, we leave this one and choose the next modulation level, which gives the lowest SF = 0.33. Again, the corresponding BER is not suitable for voice. Thereafter, we opt for the next case of SF = 0.50. The corresponding BER is suitable for voice. Thus, finally, using BER criteria

 (final) and the we select the reduced symbol interval kv−sh  constellation size Mv (final) corresponding to the SF = 0.50. It is not always needed to go to the lowest available constellation size for the minimum value of SF, which can gain a higher constellation size. Table 3 presents an example of the recursive process for choosing the final values of Mv (final)  and kv−sh (final) at a CNR of 25 dB. In this case, although one can go for Mv =4, the corresponding value of the lowest SF would be 0.50, which is higher than the lowest SF with Mv = 16. Thus, down the line, while decreasing Mv , one  has reneed not reduce further the value of Mv once kv−sh duced to 1. With this consideration, we estimate the values  of Mv (final) and kv−sh (final) for a value of CNR. Table 4 presents the final results of a shrinking operation on voice constellation for a range of CNR values with relevant param eters. The values of Mv and kv−sh at each CNR are shown  (final) in bold face corresponding to Mv (final) and kv−sh and the corresponding minimum achievable value of SF, when system BER is suitable for voice communication. The

1018

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

Fig. 2. Pseudo-code for implementing the proposed scheme.

entire scheme is presented in the form of a pseudo code in Fig. 2. Table 5 shows the final transmission parameters of the voice and data bits along with the corresponding increase in the data throughput () and SF at different values of received CNR. At each CNR, voice bits are transmitted using  Mv (final) and kv−sh (final), if the corresponding data BER permits shrinking. [As evident from Table 5, the system can achieve gain in data throughput above 15 dB of CNR, when it becomes possible to shrink the voice-symbol interval with BER  10−7 for data transmission.] The BERs for voice communication after shrinking are also shown in this table, which are calculated using kv (final) and kv (final) = log2 (Mv (final)) with the help of Eq. (12). Fig. 3 shows the plot of  vs. CNR, wherein we observe that the proposed data transmission scheme achieves

increasing benefit with an increase in CNR. As evident from Fig. 3, the percentage increase in data throughput appears as a staircase function of the received CNR. This is due to the fact that the same constellation size (an integer) has to be selected for a range of CNR values to maintain a particular BER. Therefore, the value of  remains fixed within a specific range of CNR values and increases with CNR in steps. However, the scope of improvement reduces at higher CNR, and  gradually attains a saturation. So far, we have considered the case of an integrated frame with r = Nd /Nv = (N − Nv )/Nv = 1. In Fig. 4, we present the plot of  vs. SF for different values of r. At a lower value of r, the number of free intervals after shrinking is less than that at higher value of r for a particular value of SF. Therefore, as expected,  vs. SF plots move upward with higher values of r.

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

1019

Table 5. Final transmission parameter of the proposed scheme along with the data throughput and shrinking factor Received CNR (dB)

kv (final)

 kv−sh (final)

Voice BER

Data BER

SF

%

5

0

0

No transmission

No transmission

Not applicable

Not applicable

10 12 14

2 2 3

Not applicable

7.8e−4 3.4e−5 4.4e−4

No transmission

0 0 0

0 0 0

15 17 20 22 25 28 30 32 35 40

2 2 2 3 4 5 5 6 7 8

1 1 1 1 1 1 1 1 1 1

3.4e−5 2.7e−7 7.6e−13 8.4e−7 2.6e−5 1.5e−4 3.5e−6 1.1e−4 2.8e−4 2.9e−5

9.3e−9 7.2e−13 2.5e−11 6.7e−9 1.04e−8 1.2e−8 1.5e−12 2.4e−10 2.4e−10 4.8e−28

0.50 0.50 0.33 0.33 0.25 0.20 0.20 0.16 0.14 0.12

25 25 33.3 33.3 37.5 40 40 41.6 42.8 43.7

50 45

No Dat a Transmission

40

Data Transmission

35

Δη

30 25 20 15 10 5 0 0

5

10

15 20 25 Received CNR (dB)

30

35

40

Fig. 3. Plot of percentage increase in data throughput  vs. average CNR in AWGN channel.

4. Performance in a fading channel In the previous section, the percentage increase in data throughput () is computed in a time-invariant AWGN channel, where it appears as a staircase function of received CNR. However, the scenario becomes different in the case of a fading channel. Due to the time-varying nature of the received CNR, the proposed scheme is employed in a dynamic manner with an adaptive constellation size and shrinking ratio, governed by the instantaneous value of the received CNR. Thus, the performance needs to be evaluated in terms

of the value of , averaged over all possible values of CNR for a given average CNR (¯) in a fading channel. To determine the average value of  for an average received CNR ¯ , we first split the given range of  of a fading channel into a set of discrete intervals, each corresponding to a specific Mˆ v . Specifically, we divide the given range of  into K fading regions, and define the value of  for  values falling in the ith region as i (). The regions of  and the associated  values of constellation size Mv (final) and kv−sh (final) are assumed to have been pre-computed and stored in a lookup table. The average value for  for a given value of average

1020

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

60

50 Max 256−QAM Max 256−QAM−Coste. Restricted

45 50

40

r=2

35

40

Δ ηav

Δη

30 30

25 20

20

15 10

10

5

r = 0.1

0

0 0.5

0.45

0.4

0.35

0.3 SF

0.25

0.2

0.15

0

0.1

Fig. 4. Plots of percentage increase in data throughput () vs. shrinking factor (SF) for different values of r = Nd /Nv .

5

10 15 20 25 30 Average Received CNR (dB)

35

40

Fig. 6. Plots of av vs. average CNR in Rayleigh fading channel for all constellation and square-constellation cases (maximum 256-QAM).

50 Max 256−QAM Max 64−QAM Max 16−QAM

50

40

45

35

40

30

35

25

30 Δ ηav

Δ ηav

45

20

Rayleigh Rician factor = 1 Nakagami m = 5

25

15

20

10

15

5

10 5

0 0

5

10 15 20 25 30 Average Received CNR (dB)

35

40

0

Fig. 5. Plots of av vs. average received CNR in Rayleigh fading channel for various maximum constellation sizes.

received CNR ¯ can therefore be expressed as av =

K −1    i+1 i

i

i () p() d

0

(4)

where p() is the probability density function of the received CNR for the fading channel. Fig. 5 shows the plot of av vs. ¯ of the proposed scheme in a Rayleigh fading channel wherein the maximum constellation size is limited to 256-QAM. As evident from Fig. 5, the value of av increases smoothly with ¯ (instead of staircase variation in case of a time-invariant AWGN channel, as shown in Fig. 3) and tends to attain a saturation at higher values of ¯ . The plots in Fig. 5 show the decrease in data throughput at higher CNR values if the

5

10 15 20 25 30 Average Received CNR (dB)

35

40

Fig. 7. Plot of av vs. average CNR in different fading channels.

maximum constellation size is restricted further to 64-QAM and 16-QAM. At times, a practical system may not support all feasible constellation sizes. Fig. 6 shows the comparison between the two cases, with and without constellation size restriction in a Rayleigh fading channel. In the first case, without any restriction, all feasible constellations up to 256-QAM have been considered, while in that case of restricted constellations, only square-constellations have been considered. As expected, the plots show that av is reduced with restriction is constellation size. Fig. 7 shows the plots of av vs. ¯ for an ideal case (i.e., without any restriction or upper limit on constellation size) in the presence of Rayleigh fading, Rician fading with Rician factor 1, and Nakagami-m fading with m = 5 [1].

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

It is worthwhile to mention here that in the techniques for simultaneous voice data transmission proposed in [20–23], the authors used spectrally efficient modulation for data transmission than voice transmission, whereas in our proposed technique, we use higher-order modulation for voice as compared to the modulation used for data transmission.

1021

either voice or data communication. To implement our scheme, we assume that each TDMA slot supports integrated voice and data communication, and out of 114 bits, 60 bits are used for voice (i.e., Nv = 60) and the rest i.e., 54 bits (Nd = 54) are used for data transmission (i.e., the value of r close to 1). The number of voice bits needs to be chosen with discretion. This is due to the fact that the number of bits used for voice communication should be an integral multiple of the number of bits per symbol for each modulation. In the present work, we have taken the case of Mv of 4, 8, 16, 32, and 64 with kv = 2, 3, 4, 5, and 6, respectively. Table 6 shows the number of bit intervals used for voice communication due to a shrinking operation. Initially, 60 bits are allocated for voice communication over 60-bit intervals. The system uses these bits completely for voice when CNR < 15 dB. But with CNR  15 dB, the system undertakes the data transmission, and the symbol intervals for the voice

5. Implementation in GSM-based systems In this section, we examine the proposed scheme for GSM-TDMA frames. In a GSM system, one TDMA slot is assigned to each user for either voice or data communication. Each TDMA slot consists of 156.25 bits, of which 114 bits are used for communication, wherein all the bits have the same time interval. These 114 bits are divided into two blocks of 57 bits [1]. Presently, these bits are used for

Table 6. Application of the proposed scheme in GSM-TDMA slots Received CNR (dB)

Nv

5

0

10 12 14 15 17 20 22 25 28 30 32 35 40

 Nv−sh

 Nd

Transmission status

Feasibility of shrinking

0

0

No Transmission

Not applicable

60 60 60

Not applicable

0 0 0

Voice only

Not allowed

60 60 60 60 60 60 60 60 60 60

30 30 20 20 15 12 12 10 10 10

Voice and data

Allowed

30 30 40 40 45 48 48 50 50 50

Table 7. Final transmitted bits of a GSM-TDMA slot for integrated voice and data transmission Received CNR (dB)

Nd

 Nd

Nd−sh

5

54

0

10 14

54 54

15 17 20 22 25 28 30 32 35 40

54 54 54 54 54 54 54 54 54 54

Transmission status

Feasibility of shrinking

0

No transmission

Not applicable

0 0

0 0

Voice only

Not allowed

30 30 40 40 45 48 48 50 50 50

84 84 94 94 99 102 102 104 104 104

Voice and data

Allowed

1022

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

Table 8. Analysis of the proposed scheme on the bits of a GSM-TDMA slot Received CNR (dB)

N

 Nv−sh

Nd−sh

kv (final)

 kv−sh (final)

SF



15 17 20 22 25 28 30 32 35 40

114 114 114 114 114 114 114 114 114 114

30 30 20 20 15 12 12 10 10 10

84 84 94 94 99 102 102 104 104 104

2 2 3 3 4 5 5 6 6 6

1 1 1 1 1 1 1 1 1 1

0.50 0.50 0.33 0.33 0.25 0.20 0.20 0.16 0.16 0.16

26.31 26.31 35.08 35.08 39.47 42.10 42.10 43.86 43.86 43.86

50 45

No Data Transmission

40

Data Transmission

35

Δη

30 25 20 15 10 5 0 0

5

10

15 20 25 Received CNR (dB)

30

35

40

Fig. 8. Plot of  vs. CNR in AWGN channel.

are allowed to shrink. Depending on different shrinking for different modulations (constellation size), the same number  ), of voice bits (Nv ) are sent with variable bit intervals (Nv−sh which varies from 30 down to 10 corresponding to the CNR ranging from 15 to 40 dB. Therefore, the system can send the voice using fewer bit intervals, and can thus increase the number of unused bit intervals for data transmission. These unused bit intervals (Nd = (1 − SF)Nv ) are added to the previously allocated data bits (Nd ). Hence, the number of bits available for data transmission increases to Nd−sh = Nd + (1 − SF)Nv . The number of increased bits for data transmission is given in Tables 7 and 8 along with the relevant parameters. Fig. 8 shows the variation of  in an AWGN channel with varying CNR. Using the value of  from Fig. 8, the plot of av vs. ¯ has been obtained for a Rayleigh fading channel and is shown in Fig. 9.

6. Conclusion In this paper, we have examined an integrated voice and data transmission technique, where the symbol duration of voice is shrunk with the knowledge of voice BER using adaptive modulation. In doing so, the system is able to send the same voice information with compressed symbol duration, and consequently offers some additional bits for data transmission. These additional bits for data transmission along with the predefined data bits increase the effective transmission rate. Hence, using the proposed scheme one can enhance dynamically the overall throughput in a broadband wireless network, provided the receiver CNR remains within the desirable range. In this technique, power remains invariant for voice and data. However, the shrinking of the symbol period will increase the complexity of the system at the expense of the enhanced data rate.

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

Next, we consider that the signal is modulated with a constellation size M with a symbol interval Ts = kT , where k = log2 M. If we compress the symbol interval, the bit interval is also shrunk and modified with Tsh . Therefore, the compressed symbol-interval Ts−sh is given as ksh T = kT sh . As mentioned earlier, bit energy is related to the bit interval; therefore, due to shrinking, the bit energy is also modified and can be expressed as E b−sh = A2 Tsh /2, which subsequently modified the SNR value as sh = E b−sh /. If we take the ratio between two CNR values before and after shrinking, we get

50 45 40 35 Δ ηav

30 25 20 15 10

 Eb T = = sh E b−sh Tsh

5 0

1023

(9)

40

Using (9), the modified CNR (sh ) due to shrinking can be expressed in terms of the CNR () before shrinking as

Fig. 9. Plot of av vs. average received CNR in Rayleigh fading channel.

ksh Tsh = (10) T k Finally, the BER of the M-QAM system with a compressed symbol interval as a function of modified CNR sh is approximately expressed as     2 1  BER = 1− √ (11) erfc 1.5 sh k M −1 M

0

5

10 15 20 25 30 Average Received CNR (dB)

35

The feasibility study of the proposed scheme in GSMTDMA slot indicates that the scheme can be applied to a GSM/GPRS system, by which the system will support simultaneous voice and data communication for more number of users in the present scenario.

Appendix BER for M-QAM receiver with compressed symbol interval. The BER of square M-QAM with Gray-bit mapping in a time-invariant AWGN as a function of received CNR  and constellation size M = 2k can be approximately expressed as [29]     1  2 1− √ (5) erfc 1.5 BER = k M −1 M where CNR  is given by [30] =

A2 T Eb and E b =  2

(6)

with E b being the energy per bit interval, A the amplitude of the received signal, T the bit interval, and  is the doublesided noise power spectral density. In (5), the approximation is tightest at high CNRs. This expression is, however not easily differentiable or invertible in its rate k, and so we next consider another approximation for BER, tight to within 1 dB for k  2 and BER  10−3 , given by [11]   −1.6 (7) BER ≈ 0.2 exp M −1 From (7), the constellation size M can be derived at a particular value of  as M ≈1−

1.6 log(BER/0.2)

(8)

sh = 

Substituting the value of sh from (10) into (11), we get 

  1 1.5 ×  × ksh 2 1− √ (12) erfc BER(k, ksh ) = k k(M − 1) M

References [1] Rappaport TS. Wireless communication: principle and practice. 2nd ed., Pearson Education; 2003. [2] Jakes WC. Microwave mobile communication. 2nd ed., New York: IEEE Press; 1994. [3] Cavers JK. Variable-rate transmission for Rayleigh fading channels. IEEE Trans Commun 1972; 15–22. [4] Hayes JF. Adaptive feedback communications. IEEE Trans Commun Technol 1968; 29–34. [5] Vucetic B. An adaptive coding scheme for time-varying channels. IEEE Trans Commun 1991;COM-39:653–63. [6] Webb WT, Steele R. Variable rate QAM for mobile radio. IEEE Trans Commun 1995; 2223–30. [7] Lee WCY. Estimate of channel capacity in Rayleigh fading environment. IEEE Trans Veh Technol 1990;39:187–90. [8] Kamio Y, Sampei S, Sasaoka S, Morinaga N. Performance of modulation-level-controlled adaptive-modulation under limited transmission delay time for land mobile communications. In: Proceedings of the IEEE veh technol conf (VTC95), Chicago, Illinois, 1995, p. 221–5. [9] Goldsmith AJ, Varaiya P. Capacity of fading channel with channel side information. IEEE Trans Inform Theory 1997;43(3):1986–97. [10] Ue T, Sampei S, Morinaga N. Symbol rate and modulation level controlled adaptive modulation/TDMA/TDD for personal communication systems. IEICE Trans Commun 1998;E78-B:1117–24.

1024

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

[11] Chung S, Goldsmith AJ. Degree of freedom in adaptive modulation: a unified view. IEEE Trans Commun 2001;49(9):1561–71. [12] Matsuoka H, Sampei S, Morinaga N, Kamio Y. Adaptive modulation system with variable coding rate concatenated code for high quality multi-media communication systems. IEICE Trans Commun 1996;E79-B:328–34. [13] Catreux S, Erceg V, Gesbert D, Heath Jr RW. Adaptive modulation and MIMO coding for broadband wireless data networks. IEEE Commun Mag 2002; 108–15. [14] Pecen M, Howell A. Simultaneous voice and data operation for GPRS/EDGE: class A dual transfer mode. IEEE Personal Commun 2001; 14–29. [15] Bremer G, Ko KD. Simultaneous voice and data on the general switched telephone network using framed QADM. IEEE Commun Mag 1996;34:58–63. [16] Zhang K, Pahlavan K. An integrated voice/data system for mobile indoor radio networks. IEEE Trans Veh Technol 1990;VT-39:75–82. [17] Nanda S. Analysis of packet reservation multiple access: Voice data integration for wireless networks. In: Proceedings of the IEEE global commun conference (GLOBECOM’90), San Diego, CA, December 1990. p. 1984–8. [18] Wilson N, Ganesh R, Joseph K, Raychaudhuri D. Packet CDMA versus dynamic TDMA for multiple access in an integrated voice/data PCN. IEEE J Select Areas Commun 1993;SAC-11:870–84. [19] Famolari D, Anjum F. Improving simultaneous voice and data performance in Bluetooth systems. In: Global telecommunication conference, 2002. p. 1810–4. [20] Seshadri N, Kim J. Coding and modulation for simultaneous voice and data transmission. In: Global telecommunication conference, 1995. p.17–21. [21] Alouini MS, Tang X, Goldsmith AJ. An adaptive modulation scheme for simultaneous voice and data transmission over fading channels. IEEE J Select Areas Commun 1999;17(5):837–50. [22] Hwang CS, Kim Y. An adaptive modulation method for integrated voice/data traffic over Nakagami fading channels. In: IEEE veh technol conference, 2001. p. 1649–52. [23] Hossain MdJ, Vinhaladevuni FK, Alouini MS, Bhargava VK, Goldsmith AJ. Adaptive hierarchical modulation for simultaneous voice and multi-class data transmission over fading channels. In: 4th IEEE workshop on signal processing. 2003. p. 105–9. [24] Tang X, Alouini MS, Goldsmith A. Effect of channel estimation error on M-QAM BER performance in Rayleigh fading. IEEE Trans Commun 1999;47:1856–64. [25] Paris JF, Aguayo-Torres MC, Entrambasaguas JT. Impact of channel estimation error on adaptive modulation performance in flat fading. IEEE Trans Commun 2004;52(5):716–20. [26] Oien GE, Holm H, Hole KJ. Impact of channel prediction on adaptive coded modulation performance in Rayleigh fading. IEEE Trans Veh Technol 2004;53(3):758–69. [27] Wang J, Elicin MA, Liu JCL. Multimedia support for wireless W-CDMA with dynamic spreading. Wireless Networks, vol. 8. Dordrecht: Kluwer Academic Publishers; 2002. p. 355–70. [28] Ling H, Xiaodong D, Zhimin Z, Wei D. Research on the adaptive QoS paradigm of wireless broadband applications in NGN. In: Proceedings of ICCT 2003, 2003.

[29] Proakis JG. Digital communication. 4th ed., McGraw-Hill International Edition; 2001. [30] Taub H, Schilling DL. Principles of communication systems. 2nd ed., New Delhi: Tata McGraw-Hill; 1991.

Rajarshi Mahapatra is a Research Scholar at the Department of Electronics and Electrical Communication Engineering of Indian Institute of Technology, Kharagpur. He obtained B. Tech. and M. Tech. degrees in Optics and Optoelectronics from the Department of Applied Physics, Calcutta University, Kolkata, in 1998 and 2000, respectively. His current research interests include adaptive modulation, space-time coding, and emerging techniques in wireless networks.

Anindya Sundar Dhar received his Bachelor degree in Electronics and Telecommunication Engineering from Bengal Engineering College, Sibpur, India, in 1987. In 1989, he received his M.Tech. degree in Integrated Circuits and Systems Engineering from Indian Institute of Technology, Kharagpur. He received his Ph.D. degree from the same Institute in 1994, where he is presently serving as an Assistant Professor at the Department of Electronics and Electrical Communication Engineering. His research interests include VLSI for communication and DSP architectures for real-time signal processing.

Debasish Datta received his B.Tech. degree in 1973 from the Institute of Radiophysics and Electronics, Calcutta University, and M.Tech. and Ph.D. degrees from IIT Kharagpur in 1976 and 1986, respectively. He has been engaged in teaching and research at IIT Kharagpur in the Department of Electronics and Electrical Communication Engineering for the last 28 years and currently he serves therein as the Head of the Department. During the period 1999–2002, he also served as the Chairman of G.S. Sanyal School of Telecommunications at IIT Kharagpur. In the early phase of his career, he worked for the Transmission R & D Division in Indian Telephone Industries, Bangalore, during 1976–1978, and in the Production Management Division of Audio and Intercom Systems of Philips India Ltd, Calcutta, during 1980–1981. During his stay at IIT Kharagpur, he

R. Mahapatra et al. / Int. J. Electron. Commun. (AEÜ) 63 (2009) 1012 – 1025

was awarded an Indo-US Fellowship by the Department of Science and Technology, Government of India, and the United States Agency for International development, to carry out research at Stanford University for year during 1992–1993 in the

1025

area of Coherent Optical Communications. Thereafter, he visited University of California at Davis during 1997–1999 and Chonbuk National University, South Korea, during 2003–2004 to carry out collaborative research in the area of optical networking.