Advanced Wireless Networks - 4G Technologies

138 ADAPTIVE AND RECONFIGURABLE LINK LAYER

Percentage energy savings (μ)

50

45

40

35

30

25

20

15

10

0

δ = 100 υ

δ = 10 υ

δ = 1 υ

0.2	0.4		0.6		0.8				1

Correlation coefficient channel (ρ)

Figure 4.20 Percentage energy savings vs channel burstiness when packet error rate in G state is 0 %. Performance for different choices for the delay period δ.

set equal to 2.5 times the time taken by a packet to reach the destination. If an ACK was not received before the timer expires (time-out) then the packet was considered lost and the action probabilities updated. If the chosen action was not to transmit, then a back-off timer with a time period δ was used. After the expiry of the time period δ the transmitting node again decided the next action to be generated depending on the action probabilities in {p(n)}.

Figure 4.20 shows the percentage savings in transmission energy (μ) obtained using the adaptive link layer protocol in comparison with an SW ARQ protocol for 0 % packet loss rate in the G state. Recall that the packet loss rate in the B state was assumed to be 100 %. With respect to notation from Figure 4.18, the correlation coefficient of the channel ρ for t0,0 = t1,1 is ρ = t0,0 − t0,1 = t1,1 − t1,0 is a measure of the burstiness of the channel. In these simulations the parameter δ was varied in multiples of v, where 1/v is the transmission rate. As seen from the figure the energy savings are substantial (nearly 50 %) when the correlation coefficient of the channel (ρ) is high. This is because when the correlation coefficient is high the learning algorithm is able to predict the state of the channel more accurately as it utilizes the higher ‘memory’ in the channel. When the correlation coefficient is small, the energy savings reduce (to nearly 10 %) as the behavior of the channel becomes less predictable. Nevertheless, for both high and low values of ρ the adaptive link layer protocol conserves energy when compared with the traditional ARQ scheme.

The tradeoff in higher energy savings is an increased mean delay per packet transmission, as seen in Figure 4.21. We observe that the difference in additional mean packet delay (w.r.t. the SW ARQ) between δ = v and δ = 10v is reasonably small. We reiterate that δ is a user-defined parameter which can be tuned to a specific application’s mean delay requirements. Finally, we note that similar performances can be observed when the proposed method is also compared with other traditional ARQ protocols.

INFRARED LINK ACCESS PROTOCOL

139

	1.8	x 10–4
	1.6					δ = 100 υ
						δ = 10	υ
	1.4					δ = 1 υ
) (s)
	1.2
(
packet	1
per
delay	0.8
Mean	0.6
	0.4
	0.2
	0
		0	0.2	0.4	0.6	0.8	1

Correlation coefficient channel (ρ)

Figure 4.21 Mean delay per packet vs channel burstiness when G state packet loss rate is equal to 0 %.

4.5 INFRARED LINK ACCESS PROTOCOL

Recent growth in laptop computers and on portable devices, such as personal digital assistants (PDAs) and digital cameras, has led to an increasing demand for information transfer from or between portable devices [52]. Digital representation of information is expanding to new devices such as video and photocameras. New devices have ‘computer-like’ capabilities for storing and retrieving information such as mobile phones and portable information gathering appliances. Computer manufacturers have adopted the Infrared Data Association (IrDA) standard [53] and almost every portable computer and all Windows CE device on the market today contains an infrared (IR) port according to standards developed by IrDA. Laptop computers, PDAs, digital cameras, mobile phones and printers are examples of devises with IrDA links. More than 40 000 000 devices are shipped each year with IrDA ports [54] capable of using the unregulated IR spectrum for their cable-less communication needs. The IrDA standard addresses low-cost, indoor, short-range, half duplex, point-to- point links [55]. The IrDA physical layer specification (Ir-PHY) [56] supports optical links from 0 to at least 1 m, an angle of ±15◦ at a BER of less that 10−8. Ir-PHY version 1.0 serial IR (SIR) specification [57] supported data rates up to 115.2 kb/s using standard serial hardware, Ir-PHY version 1.1 fast IR (FIR) [58] extended the data rate to 4 Mb/s, and fi- nally Ir-PHY version 1.3 very fast IR (VFIR) [59] specification added the 16 Mb/s link rate. The IrDA hardware is controlled by a link-layer protocol, the IrDA link access protocol (IrLAP) [60, 61]. IrLAP is based on the widely used high-level data link control (HDLC) protocol operating in NRM (see Chapter 1). The performance of IrDA optical wireless links may be measured by the throughput which can be drawn at the IrLAP layer. In Vitsas and Boucouvalas [62, 63] performance evaluation of the IrLAP protocol and deriving optimum values for link layer parameters for maximizing throughput is presented. In the literature, a

140 ADAPTIVE AND RECONFIGURABLE LINK LAYER

mathematical model for the IrLAP throughput using the concept of a frame’s ‘virtual transmission time’ is presented [52, 53] based on the HDLC analysis model presented in Williams [54]. However, this model does not lead to a simple formula for the IrLAP throughput. In Vitsas and Boucouvalas [57, 58], a new mathematical model using the average window transmission time (WTT) is developed. By taking advantage of IrLAP half-duplex operation, this model leads to a simple closed form formula for IrLAP throughput. The formula relates throughput to physical layer parameters, such as link BER, link data rate and minimum turnaround time, and with link layer parameters such as frame size window size and frame overhead. As this equation gives an intuitive understanding of the performance of IrDA links, it is very valuable for designers and implementers of such links. By setting the first derivative equal to zero, the optimum values for window size and frame length that maximize throughput are derived. Formulas for all IrLAP time-consuming tasks are also presented, allowing evaluation of link parameter values to throughput performance. IrLAP performance is examined for various link parameters, such as BER, data rate and window size and compared with optimum performance achieved using optimum window size and frame length values. Optimum window and frame-size values can be easily implemented and result in significant throughput increase, especially for links experiencing high BERs. However, implementing optimum frame-size values on retransmissions requires buffer reorganization at a low level.

4.5.1 The IrLAP layer

IrLAP is the IrDA data link layer. It is designed based on the pre-existing HDLC and synchronous data link control protocols (see Chapter 1, also Reference [60]). IrLAP stations operate in two modes: the normal disconnect mode during the contention period and the NRM during the connection period. In the contention period, a station advertises its existence to the neighboring stations along with the link parameters it supports and wishes to employ during the connection period. One of the participating stations becomes the primary station. Any station may claim to become the primary station but, at the end of the contention period, only one station is granted the primary role and all other stations are assigned the secondary role. All data traffic during the connection period is sent to or from the primary station. A secondary station wishing to communicate to another secondary station does so through the primary station. The parameters negotiated and agreed on during the contention period are given below:

(1)Data rate (C) – this parameter specifies the station’s transmission rate.

(2)Maximum turnaround time (Tmax) – this parameter specifies the maximum time interval a station can hold transmission control. For data rates less than 115.2 kb/s, the maximum turnaround time must be 500 ms. A smaller value may be agreed between the two stations for 115.2 kb/s or higher data rates.

(3)Data size (l) – this is the maximum length allowed for the data field in any received information frame (I frame). This parameter has an upper limit of 2048 bytes (16 kb).

(4)Window size (Wmax) – this is the maximum number of unacknowledged frames I a station can receive before it has to acknowledge the number of frames received

INFRARED LINK ACCESS PROTOCOL

141

Nr (3 b) P/F Ns (2 b) Information Frame

Nr (3 b)			P/F		X (2 b)
	Supervisory Frame
X (3 b)			P/F		X (2 b)
	Unnumbered Frame

0

01

11

Address (7 b)

C/R

Control (8 b)

Information (0 to 16384 b)

IrLAP Packet

START (8 b)

IrLAP packet (N b)

FCS (16/32 b)

STOP (8 b)

Physical Layer Frame

Figure 4.22 IrDA SIR and FIR frame structure.

correctly. An acknowledgement may be requested by the transmitting station before the window size is reached. This parameter has an upper limit of 7 for data rates up to 4 Mb/s and 127 for 4 and 16 Mb/s [60, 61].

(5)Minimum turnaround time (tta) – this is the time required by the station’s receive circuit to recover after the end of a transmission initiated from the same station (turnaround latency). Each station must wait a minimum turnaround time delay when moving from receive mode to transmit mode to ensure that the receive circuit of the station that was transmitting is given enough time to recover. This is the time required to change link direction.

Both stations must use the same data rate. However, parameters (2)–(5) are negotiated and agreed independently for each station. The IrDA frame structure is shown in Figure 4.22. The FCS contains a 16 b CRC for data rates up to 4 Mb/s and a 32 b CRC for 4 Mb/s and higher rates. IrLAP employs the following frame types:

(1)Unnumbered frames (U frames) are used for link management. ‘U frame’ functions include discovering and initializing secondary stations, reporting procedural errors not recoverable by retransmissions, etc.

(2)I frames carry information data across the link during the connection period. The I -frame control field contains send and receive frame counts to ensure ordered frame reception.

(3)Supervisory frames (S frames) assist in information data transfer, although S frames never carry information data themselves. They are used to acknowledge correctly received frames, request an acknowledgement from the communicating station, con-

vey station conditions, etc.

142 ADAPTIVE AND RECONFIGURABLE LINK LAYER

The control field contains an identifier, which determines the frame type. Depending on frame type, the control field may contain a send sequence number Ns, used to number the transmitted frames. It may also contain a receive sequence number Nr, used to indicate the expected sequence number of the next I frame. SIR and FIR specifications employ an 8 b-long control field (Figure 4.22). Ns and Nr occupy 3 b each in the control field, thus, Ns and Nr cycle through values from 0 to 7 and maximum window size is 7. Very fast infrared (VFIR) specification extended the control field to 16 b for the 4 and 16 Mb/s data rate IrDA links. In this case, Ns and Nr occupy 7 b each, cycling through values 0 to 127 and a maximum window size of 127 is allowed.

Within the control field, the P/F bit implements token passing between stations. When it is set by the primary station, it is the poll (P) bit. When it is set by the secondary station, it is the final (F) bit. The primary uses the P bit to reverse link direction and solicit a response from the secondary. The secondary responds by transmitting one or more frames and by setting the F bit of the last frame it transmits, thus, reversing link direction and returning transmission control to the primary. IrLAP primary and secondary stations also employ the P timer. P timer is assigned with the maximum turnaround time (Tmax) agreed between stations during the contention period and represents the maximum time a station can hold transmission control. Each station starts the P timer upon reception of a frame with the P/F bit set and stops the P timer when it transmits a frame with the P/F bit set. If the P timer expires, meaning that the station holds transmission control longer than allowed, the station immediately sends a receive ready (RR) S frame with the P/F bit set to pass transmission control. The primary station also employs an F timer to limit the time a secondary station can hold transmission control. The primary starts the F timer upon transmission of a frame with the P bit set and stops the F timer upon reception of a frame with the F bit set. F-timer expiration means that the secondary failed to return transmission control within the agreed time period. Since the secondary’s P -timer operation guarantees that this never happens, F-timer expiration can only be explained by the loss of either the frame containing the P bit or the frame containing the F bit. The primary resolves this situation by transmitting an RR frame with the P bit set when theF timer expires.

4.5.2 IrLAP functional model description

In this section, the transmission of a large amount of information data from the primary to the secondary station is considered. The saturation case is assumed, where the primary station always has information data ready for transmission.

The parameters used in the model are shown in Table 4.4. In the contention period, the primary station determines the window size N it will employ. N represents the maximum number of I frames the primary can transmit before soliciting an acknowledgement. The maximum window-size parameter Wmax is negotiated and agreed between the two stations during the contention period. However, the maximum time a station can hold transmission control Tmax must always be obeyed and, according to IrLAP specification [60], Tmax combined with frame length and link data rate may limit the window size applied. In other words, if the time needed for transmitting Wmax frames carrying ‘frame length’ information bytes at the link data rate exceeds Tmax, then a smaller window size must be employed. Thus, N is given by

N = min {Wmax, floor(Tmax/tI )}

(4.43)

	INFRARED LINK ACCESS PROTOCOL			143
	Table 4.4 System parameters
Parameter	Description	Unit
C	Link data bit rate	b/s
pb	Link bit error rate	—
p	Frame error probability	—
l	I-frame length data length	b
l	S-frame length/I-frame overhead	b
tI	Transmission time of an I-frame	s
tImax	Transmission time of an I-frame with
	16 k bits user data	s
tS	Transmission time of an S-frame	s
tta	Minimum turn-around time	s
tack	Acknowledgement time	s
Tmax	Maximum turn-around time	s
tFout	F-timer time-out period	s
Wmax	Maximum window size	Frames
N	Window size	Frames
Df	Frame throughput	Frame/s
Db	Data throughput	bits/s

where min is ‘the lesser of’ and floor is ‘the largest integer not exceeding’. In this section, Tmax is always fixed to 500 ms. The information transfer procedure used in the model is presented in Figure 4.23. Each node holds three variables, Vs for counting I frames transmitted, Vr for counting I frames received, and w indicating the number of the remaining I frames the station can transmit before reversing link direction. The primary also employs an F timer for limiting the secondary’s transmission period. When the primary station sends an I frame, the Ns subfield of the frame’s control field is assigned the current Vs value and is increased by one (modulo 8 or 128 depending of the control field size employed). The primary also makes a buffer copy of the frame for possible retransmissions. Since the primary always has information ready for transmission, it immediately checks the w value. If w is not equal to 1, the primary reduces w by 1, transmits the I frame with the P bit not set and the actions previously described are repeated. When w reaches 1, indicating that the next frame should be the last frame in the window transmission, the primary sets the bit P to poll the secondary and transmits the I frame. The primary also assigns N to w for the next N window frame transmission and starts the F timer.

When the secondary station receives an I frame, it compares the received frame sequence Ns value with the expected Vr value. If Ns equals Vr (the received frame is in sequence), Vr is increased by 1 (modulo 8 or 128) and information data is extracted and passed to the upper layer. If the received frame is not in sequence (one of the previous I frames in current window transmission was lost due to a CRC error), the frame is discarded and Vr remains unchanged. The secondary station also checks the P bit. If the P bit is set and, as the current model assumes that the secondary station never has information for transmission, it awaits a minimum turnaround time tta to allow for the hardware recovery latency and transmits an S frame with the F bit set. The S frame’s Nr field contains Vr, a value informing the primary

144 ADAPTIVE AND RECONFIGURABLE LINK LAYER

Primary

< get new or buffered data for transmission >;

Ns = Vs ;

Vs = (Vs + 1) mod (8 or 128); < make buffer copy >;

w = w-1; Do not set P bit; transmit I-frame;

No

w = 1 ?

Yes

w = N; Set P bit;

transmit I-frame; start F-frame;

Wait to	F-timer
receive	Expires

S-frame received

Compare Nr with Vs and determine buffered frames to be retransmitted;

Secondary

S-frame received

	Wait
	I-frame received
	≠
	Ns = Vr ?
	=
				Discard
	Vr = (Vr + 1) mod (8 or 128);
				frame;
	extract and pass data to
	upper layer;
	Set			Wait tta ;
	P bit ?			Nr = Vr ;
			transmit S-frame;
	Not set

Transmit S-

frame;

Figure 4.23 Information transfer procedure.

REFERENCES 145

of the number of I frames received correctly and in sequence in the previous window transmission. When the primary receives the frame, it resumes I -frame transmission as transmission control was returned to the primary by means of the F bit. The primary first compares the received S frame’s Nr with its current Vs value. If Nr equals Vs (all frames in the previous window transmission were received correctly by the secondary), the primary transmits I frames containing new information data to the secondary. If Nr is not equal to Vs, one or more I frames in the previous window transmission are lost. The primary retransmits buffered frames starting from the indicated Nr position before new data can be transmitted.

If the last I frame that contains the P bit is lost, the secondary station fails to respond as it does not realize that it has transmission control. The situation is resolved by the primary’s F-timer expiration. The primary realizes that the secondary failed to respond within the agreed time period and transmits an S frame forcing the secondary to respond. In the current model, S frames are considered small enough to always be received error-free.

The saturation case model considered in this section can be summarized as follows. The transmitting station always has information ready for transmission. As a result, it transmits a window of N consecutive I frames and reverses the link direction by setting the P bit in the last I frame. The receiver awaits a minimum turnaround time and responds with an RR S frame indicating the expected sequence number of the next frame. RR frames always have the F bit set. The transmitter determines the number of frames correctly received before any error(s) occurred and repeats the erroneous frame and the frames following it, in the next window, followed by new frames to form a complete N frame transmission. If the last frame in a window transmission is lost, the receiver fails to respond as the P bit is lost. When F timer expires, the primary station sends an RR S frame with the P bit set forcing the secondary station to acknowledge correctly received frames. Additional discussion on infrared link access protocol can be found in References [62–65].

REFERENCES

[1]U. Chelentano and S. Glisic, Effective link layer capacity, in PIMRC05, Berlin, 12–14 September 2005, paper I06-2.

[2]D. Wu and R. Negi, Effective capacity: a wireless link model for support of quality of service, IEEE Trans. Wireless Commun., vol. 2, no. 4, 2003, pp. 630–643.

[3]G. Liebl, T. Stockhammer and F. Burkert, Modelling and simulation of wireless packet erasure channels, in Proc. Tech/MRPG Symp. Wireless Personal Commun., Blacksburg, VA, May 2000, pp. 203–214.

[4]T. Rancurel, D. Roviras, J. Conan and F. Castanie, Expression of the capacity for the Gilbert channel in presence of interleaving, Proc. Int. Conf. Acoustics Speech and Signal Processing (ICASSP), vol. 4, 2000, pp. 2533–2536.

[5]T. Rancurel, D. Roviras, J. Conan and F. Castanie, Effect of interleaving on a Markov channel, Proc. Int. Conf. Acoustics Speech and Signal Processing (ICASSP), vol. 5, 2000, pp. 2661–2664.

[6]C. Pimentel and I.F. Blake, Modeling burst channels using partitioned Fritchman’s Markov models, IEEE Trans. Vehicle Technol., vol. 47, no. 3, 1998, pp. 885–899.

[7]A. Willig, A new class of packetand bit-level models for wireless channels, in Proc. IEEE Symp. Personal Indoor Mobile Radio Communications, Lisboa, September 2002.

146ADAPTIVE AND RECONFIGURABLE LINK LAYER

[8]J.N.L. Brummer,¨ Characterization of digital channels using hidden Markov models,

Proc. Symp. Communication Signal Processes, 11 September 1992, pp. 183–188.

[9]H.S. Wang and N. Moayeri, Finite-state Markov channel – a useful model for radio communication channels, IEEE Trans. Vehicle Technol., vol. 44, no. 1, 1995, pp. 163– 171.

[10]H. Steffan, Adaptive generative radio channel models, in Proc. IEEE Symp. Personal Indoor Mobile Radio Communications, vol. 1, 1994, pp. 268–273.

[11]C.-D. Iskander and P.T. Mathiopoulos, Analytical level crossing rates and average fade durations for diversity techniques in Nakagami fading channels, IEEE Trans. Commun., vol. 50, no. 8, 2002, pp. 1301–1309.

[12]F. Babich and G. Lombardi, A Markov model for the mobile propagation channel, IEEE Trans. Vehicle Technol., vol. 49, no. 1, 2000, pp. 63–73.

[13]M. Zorzi and R.R. Rao and L.B. Milstein, ARQ error control for fading mobile radio channels, IEEE Trans. Vehicle Technol., vol. 46, no. 2, 1997, pp. 445–455.

[14]H.S. Wang, On verifying the first-order Markovian assumption for a Rayleigh fading channel model, in Proc. IEEE Int. Conf. Universal Personal Commun., 1994, pp. 160– 164.

[15]H.S. Wang and P.-C. Chang, On verifying the first-order Markovian assumption for a Rayleigh fading channel model, IEEE Trans. Vehicle Technol., vol. 45, no. 2, 1996,

pp.353–357.

[16]A. Abdi, Comments on verifying the first-order Markovian assumption for a Rayleigh fading channel model, IEEE Trans. Vehicle Technol., vol. 48, no. 5, 1999, p. 1739.

[17]C.C. Tan and N.C. Beaulieu, On first-order Markov modeling for the Rayleigh fading channel, IEEE Trans. Commun., vol. 48, no. 12, 2000, pp. 2032–2040.

[18]F. Babich and G. Lombardi, On verifying a first-order Markovian model for the multithreshold success/failure process for Rayleigh channel, Proc. IEEE Symp. Personal Indoor Mobile Radio Communications, vol. 1, 1997, pp. 12–16.

[19]D. Qiao and S. Choi, Goodput enhancement in IEEE802.11a wireless LAN via link adaptation, in Proc. IEEE Int. Conf. Commun., Helsinki, June 2001.

[20]N.C. Ericsson, Adaptive modulation and scheduling of IP traffic over fading channels, in Proc. IEEE Vehicle Technology Conf. (VTC’99), Amsterdam, vol. 2, 19–22 September 1999, pp. 849–853.

[21]A.J. Goldsmith and S.-G. Chua, Variable-rate variable-power MQAM for fading channels, IEEE Trans. Commun., vol. 45, no. 10, 1997, pp. 1218–1230.

[22]S. Vishwanath and A. Goldsmith, Adaptive turbo-coded modulation for flat-fading channels, IEEE Trans. Commun., vol. 51, no. 6, 2003, pp. 964–972.

[23]V. Tralli and M. Zorzi, Markov models for the physical layer block error process in

aWCDMA cellular system, Proc. IEEE Global Telecommun. Conf., vol. 2, 17–22 November 2002, pp. 1925–1929.

[24]A.J. Paulraj, D.A. Gore, R.U. Nabar and H. Bolcskei,¨ An Overview of MIMO communications – A key to gigabit wireless, IEEE Proc., vol. 92, no. 2, 2004, pp. 198–218.

[25]S.-T. Chung and A.J. Goldsmith, Degrees of freedom in adaptive modulation: a unified view, IEEE Trans. Commun., vol. 49, no. 9, 2001, pp. 1561–1571.

[26]L. Hanzo, C.H. Wong and M.S. Yee, Adaptive Wireless Transceivers. Wiley: Chichester, 2002, pp. 191–255.

[27]Y.Y. Kim and S.-Q. Li, Capturing important statistics of a fading/shadowing channel for network performance analysis, IEEE J. Select. Areas Commun., vol. 17, no. 5, 1999, pp. 888–901.

REFERENCES 147

[28]D. Haccoun and G. Begin, High-rate punctured convolutional codes for Viterbi and sequential decoding, IEEE Trans. Commun., vol. 37, no. 11, 1989, pp. 1113–1125.

[29]S. Wei and D.L. Goeckel, Error statistics for average power measurements in wireless communication systems, IEEE Trans. Commun., vol. 50, no. 9, 2002, pp. 1535–1546.

[30]J.G. Proakis, Digital Communications, 2nd edn. McGraw-Hill: Singapore, 1989.

[31]N. Wax (ed.), Selected Papers on Noise and Stochastic Processes. Dover: New York, 1954, pp. 133–294.

[32]M. P¨atzold and F. Laue, Level-crossing rate and average duration of fades of deterministic simulation models for rice fading channels, IEEE Trans. Vehicle Technol., vol. 48, no. 4, 1999, pp. 1121–1129.

[33]N.C. Beaulieu and X. Dong, Level crossing rate and average fade duration of MRC and EGC diversity in Ricean fading, IEEE Trans. Commun., vol. 51, no. 5, 2003, pp. 722–726.

[34]A.M. Law and W.D. Kelton, Simulation Modeling and Analysis. Industrial Engineering Series, 2nd edn. McGraw-Hill: Singapore, 1991, pp. 372–416.

[35]F.J. Block and M.B. Pursley, A protocol for adaptive transmission in direct-sequence spread-spectrum packet radio networks, IEEE Trans. Commun., vol. 52, no. 8, 2004, pp. 1388–1397.

[36]P. Lettieri and M.B. Srivastava, Adaptive frame length control for improving wireless link throughput, range, and energy efficiency, in Proc. IEEE INFOCOM’98, 29 March to 2 April 1998, pp. 564–571.

[37]I. Joe, An adaptive hybrid ARQ scheme with concatenated FEC codes for wireless ATM, in Proc. ACM/IEEE MobiCom’97, 1997, pp. 131–138.

[38]D.A. Eckhardt and P. Steenkiste, Improving wireless LAN performance via adaptive local error control, in Proc. IEEE ICNP’98, 1998, pp. 327–338.

[39]M. Elaoud and P. Ramanathan, Adaptive use of error-correcting codes for real-time communication in wireless networks, Proc. IEEE INFOCOM’98, 1998, pp. 548–555.

[40]S. Choi and K.G. Shin, A class of adaptive hybrid ARQ schemes for wireless links, IEEE Trans. Vehicular Technology, vol. 50, no. 3, May 2001, pp. 777–790.

[41]T.S. Rappaport, Wireless Communications: Principle and Practice. Prentice-Hall: Englewood Cliffs, NJ, 1996.

[42]P. Lettieri, C. Fragouli and M.B. Srivastava, Low power error control for wireless links, in Proc. ACM/IEEE MobiCom’97, 1997, pp. 139–150.

[43]Y.H. Lee and S.W. Kim, Adaptive data transmission scheme for DS/SSMA system in bandlimited Rayleigh fading channel, in Proc. IEEE ICUPC’95, 1995, pp. 251–255.

[44]M. Zorzi and R.R. Rao, Error control and energy consumption in communications for nomadic computing, IEEE Trans. Comput. (Special Issue On Mobile Computing), vol. 46, 1997, pp. 279–289.

[45]P. Lettieri, C. Schurgers and M.B. Srivastava, Adaptive link layer strategies for energy efficient wireless networking, Wireless Networks, vol. 5, 1999, pp. 339–355.

[46]M. Zorzi and R.R. Rao, Energy constrained error control for wireless channels, IEEE Person. Commun., vol. 4, 1997, pp. 27–33.

[47]C. Chien, M. Srivastava, R. Jain, P. Lettieri, V. Aggarwal and R. Sternowski, Adaptive radio for multimedia wireless links, IEEE JSAC, vol. 17, no. 5, 1999, pp. 793–813.

[48]K. Narendra and M. Thathachar, Learning Automata: an Introduction. Prentice Hall: Englewood Cliffs, NJ, 1989.

[49]M. Norman, Some convergence theorems for stochastic learning models with distance diminishing operators, J. Math. Psychol., vol. 5, 1968, pp. 61–101.