Sie sind auf Seite 1von 9

Wireless Networks 7, 487495, 2001 2001 Kluwer Academic Publishers. Manufactured in The Netherlands.

An Adaptive-Transmission Protocol for Frequency-Hop Wireless Communication Networks


JOHN H. GASS, JR.
ITT Industries Aerospace/Communications Division, Fort Wayne, IN, USA

MICHAEL B. PURSLEY, HARLAN B. RUSSELL and JEFFREY S. WYSOCARSKI


Department of Electrical and Computer Engineering, Clemson University, Clemson, SC, USA

Abstract. An energy-efcient adaptive-transmission protocol for mobile frequency-hop spread-spectrum wireless communication networks is described and evaluated. The purpose of the protocol is to permit each of the mobile terminals to adjust its transmitter power and code rate to match the characteristics of the time-varying communication links in the network. The proposed adaptive-transmission protocol bases its choice of transmission parameters on a very simple form of side information that is easy to obtain in a FH communication receiver. The performance of the adaptive-transmission protocol is evaluated for networks in which each communication link may have a time-varying propagation loss and intermittent partial-band interference. Our results demonstrate that the adaptive-transmission protocol can improve the utility of a link and reduce energy consumption by adjusting the transmission parameters in response to changes in the side information. Keywords: adaptive-transmission protocol, frequency-hop spread spectrum, wireless network protocols, mobile communications

1. Introduction The quality of a communication link in a mobile wireless communication network varies from one transmission to the next, due in part to the mobility of the communication terminals and the dynamic environment in which they must operate. As terminals move, the changes in their relative positions produce variations in propagation losses on the communication links. The motion of the terminals may also result in changes in the interference experienced by the communication receivers in the terminals, and the interference may uctuate, even for stationary terminals. A common approach for the design of the transmission system is to choose the transmission parameters to provide adequate performance for the worst channel conditions for which reliable communication is required. Such a transmission system makes poor use of energy and transmission time when channel conditions are more favorable. A properly designed adaptive-transmission protocol can also provide reliable communication over the worst channels, but it can adjust the transmission parameters to reduce energy and shorten the transmission time when the channel conditions are more favorable. The interaction between the physical and network layers of a protocol suite has previously been exploited in adaptiveforwarding and adaptive-routing protocols. Specically, side information from the demodulator and decoder in frequencyhop receivers is employed in [1] to give a quantitative measure of the capabilities of the links for use in adaptive rout This research was supported by the US Army Research Ofce under

grant DAAD19-99-1-0289, and by the US Ofce of Naval Research under grants N00014-96-1-0869, N00014-97-1-0776, and N00014-00-1-0565.

ing. Although adaptive routing improves the performance of a network with time-varying links (e.g., by avoiding poorquality links), even better performance can be achieved by adapting the transmission parameters to improve the utility and efciency of the links. The adaptive-transmission protocol should be employed to make a communication link as useful and efcient as possible, and then the adaptive-routing protocol can decide whether to use that link in one or more of the routes that it needs. The ability of the adaptive-transmission protocol to maintain the utility of a communication link as the propagation loss and interference increase is of particular importance for real-time trafc, such as voice communications, because such trafc is often sent over a route that is set up at the beginning of each session (e.g., a wireless virtual circuit). It is then very important to keep each link in this route functioning adequately for the duration of the session. Adaptivetransmission methods can help maintain the links in a route until a new connection can be established. Our general approach to adaptive transmission can be used to adapt the transmitter power, code rate, symbol rate, and other transmission parameters. The specic adaptivetransmission protocol that is dened and evaluated in this paper adjusts the power and code rate for transmissions in a frequency-hop spread-spectrum packet communication network. Its performance is evaluated for channels in which the propagation loss and partial-band interference are allowed to vary from one transmission to the next. The adaptation on a given link is based on side information that consists of counts of errors and erasures that are obtained from previous transmissions on that link. In particular, no power measurements are required. The side in-

488

GASS ET AL.

formation used in the adaptive-transmission protocol is the same as the side information that is used for adaptive routing in [1]. In fact, our adaptive-transmission protocol requires no additional overhead beyond what is used in this adaptiverouting protocol. The protocol described in this paper is based in part on previous work on adaptation of code rate [2] and adaptation of power and code rate in frequency-hop communication systems [35]. The primary difference between the adaptivetransmission protocol described and evaluated in this paper and the protocols of [35] is the way in which the power is adapted. In our protocol the adaptation of the power employs an estimate of the block error probability for the next transmission. This estimate is derived from side information obtained in the previous transmission.

2. Features of the frequency-hop network Some terminology is needed to distinguish among the several different types of packets that we must consider. An information packet is a block of information bits that is to be sent from one terminal to another. Each information packet is encoded into a data packet for transmission over a communication link. In this paper the term packet refers to a data packet unless specied otherwise. Other types of packets are dened as needed. The model for frequency-hop transmission is described in [6]. Each packet consists of multiple code words from an (n, k) ReedSolomon code that are interleaved and transmitted during an interval that spans several consecutive dwell intervals. In this paper we consider the extended Reed Solomon codes, so the block length n is a power of 2, and the alphabet size for the code is also n. The rate of the code is r = k/n. The value of n is xed, so the adaptivetransmission protocol changes the rate of the code by changing the value of k. The system may employ binary signaling, in which case the nonbinary symbols for the ReedSolomon code consist of sequences of log2 n binary symbols [6]. Alternatively, any alphabet of size n may be used. Each packet consists of one or more error-control blocks. An errorcontrol block is a set of L code words that are interleaved and transmitted in n consecutive dwell intervals in such a way that each code word has exactly one code symbol in each dwell interval [6]. The variations in the channel are such that the path loss may change from one transmission to the next. In addition, partial-band interference may affect some transmissions and not others. Changes in the path loss are included in the model to account for variations in the communication range and effects of short-term shadowing. Partial-band interference consists of frequency-hop multiple-access interference from within the network and band-limited interference from external sources. The interference from external sources is modeled as band-limited white Gaussian noise that occupies a fraction of the band. Its one-sided power spectral density is 1 NI .

Full-band white Gaussian noise with one-sided power spectral density N0 is included in the model, primarily to account for thermal noise in the receivers. The total power spectral density is 1 NI + N0 in a fraction of the band and N0 in the remainder of the band. The test-symbol method [6] is used to provide side information for making erasures at the decoder input, although several other methods are also suitable, such as those that use parity checks [7], Viterbis ratio-threshold statistic [8], or Bayesian decision theory [9]. The side information need not be perfect in any sense, and the only side information that is employed is generated in the receiver by counting errors and erasures. Neither the receiver nor the transmitter are provided any other information about the state of the channel. In the test-symbol method, binary test symbols are transmitted in each dwell interval in addition to the encoded data. These symbols are known to both the transmitter and receiver, so the receiver can determine which test symbols are in error at the demodulator output. Each code symbol in a dwell interval is erased if the number of errors among the test symbols in that dwell interval exceeds a prescribed threshold. The erasures replace the corresponding code symbols at the demodulator output before errors-and-erasures decoding is applied to the received packet. For each code word in a packet, there is a corresponding received word at the input to the errors-and-erasures decoder. Each received word has a combination of erasures, erroneous symbols, and correct symbols. The number of erasures in each received word is known. The number of errors in a received word is known if that received word decodes correctly, and it can be estimated even if the word does not decode correctly [1]. A packet is said to be received correctly if each of its received words is decoded correctly.

3. The adaptive-transmission protocol In this section and the next we describe an adaptivetransmission protocol for a mobile frequency-hop spreadspectrum wireless communication network. Most of the features of the protocol that are independent of the channelaccess protocol are described in this section. The elements of the adaptive-transmission protocol that depend on the channel-access protocol are described in the next section for a network that employs a particular reservation-based method for channel access. The transmission parameters that are adapted are the rate of the error-control code and the amount of power that is used to transmit a packet. The goal is to conserve energy by transmitting at low power and using a high-rate code whenever the link conditions permit. It is desirable that the code rate be decreased or the power level increased only when necessary to overcome increased interference or propagation loss. A neighbor of a particular terminal is any other terminal in the network that is within communication range. Consider

ADAPTIVE-TRANSMISSION PROTOCOL FOR FREQUENCY-HOP NETWORKS

489

the scenario in which terminal 1 wishes to transmit a packet to terminal 2, which is one of its neighbors. In this paper we are not concerned with whether terminal 1 originated the packet or is simply relaying it to terminal 2, nor are we concerned with whether terminal 2 wants to use the information in the packet or intends only to relay it to another terminal. We are interested only in the transmission of the packet from terminal 1 to terminal 2, so it is convenient to refer to terminal 1 as the source for the transmission and terminal 2 as the destination for the transmission. Terminal 1 is not necessarily the original source of the information represented by the packet, and terminal 2 is not necessarily the packets nal destination in the network. Upon receipt of a packet, the destination determines which dwell intervals in the packet should be erased. The average number of dwell intervals erased per error-control block in the packet is denoted by e and referred to as the erasure count. Because of the format used for the packet, the number of erasures is the same for each received word in an error-control block. If the packet has only one error-control block, e is just the number of dwell intervals that are erased in the packet. If there are multiple error-control blocks in the packet, e is not necessarily an integer. Regardless of the number of error-control blocks per packet, e represents the average number of erasures per received word. If a received word in the packet is decoded correctly (e.g., as veried by an additional error-detection code), the number of symbol errors in the received word is determined from the decoder or its output. If the decoding procedure does not supply this information, the number of errors can be found by encoding the decoder output and comparing the resulting code word to the received word. If the received word does not decode correctly, or if there are too many erasures to attempt a decoding, the number of symbol errors is not known, but it can be estimated as described in [1]. If more than n k dwell intervals are erased, the estimate of the number of symbol errors is set to 0. The error count, denoted by t, is the average number of errors per word in the received packet. If a packet is received correctly, the destination sends an acknowledgment packet, referred to as an ACK, to the source. If the packet is not received correctly and the destination is able to determine the source for the packet transmission, the destination sends the source a negative acknowledgment packet, referred to as a NACK. The destination includes in each ACK and NACK the appropriate information derived from the counts of errors and erasures for the most recent packet from the source. This information may consist of the counts themselves, or it may be the specications for the power level and code rate that are derived from the error and erasure counts. The information returned in the ACK or NACK is employed at the source to adapt the transmission parameters for the next packet terminal 1 sends to terminal 2. It is possible that an ACK or a NACK is sent by the destination but not received by the source. The protocols mechanism for adapting the transmission parameters

in the absence of one or more ACKs or NACKs is described later in this section. The method described in this paper for adapting the power level is different than the methods employed in [35], but all of these methods have the common feature that the adaptation of the code rate and power level depends only on the error and erasure counts. The selection of the code rate for the next packet transmission from the source to the destination is determined by a comparison of the values of the erasure count e and the adaptation parameter . If e > the code rate is decreased if possible (i.e., if the code rate is not already at its lowest value), but if e the code rate is increased if possible. This part of the adaptive-transmission algorithm is the same as in [5], and it is based on the approach described in [2,4]. Alternative methods for adapting the code rate are given in [3]. Once the code is selected, the power level is determined for use in the next packet transmission. The adaptation of the power level is based on the parameters of the code that will be used for the next transmission and on a statistic that is derived from the error and erasure counts from the previous reception. This statistic is an estimate of the probability of error for an error-control block in the next transmission. This approach for adapting the power differs signicantly from the power adaptation methods used in [35]. Suppose an (n, k) code is selected for the next transmission. The values of n, k, t, and e are used to determine the statistic Pb (e, t, l) that is dened by Pb (e, t, l) = 1 1 Pw (e, t, l) , where Pw (e, t, l)
ne L

(1)

=
i= (le)/2

ne i

pi (1 p)

ne i

(2)

p = t/(n e), and l = n k. The probability Pw (e, t, l) is an estimate of the probability that a given received word in the next packet transmission will not decode correctly if the redundancy of the code for the next transmission is l and if the erasure count and error count for the previous packet are e and t, respectively. The probability p is an estimate of the probability of error for an unerased code symbol in the next transmission. The statistic Pb (e, t, l) is compared with two adaptation parameters 1 and 2 that satisfy 1 < 2 . If Pb (e, t, l) 1 the power is decreased if possible. If 1 < Pb (e, t, l) 2 the power level is unchanged. Otherwise, the power is increased if possible. There are a number of scenarios in which the source may not receive either an ACK or a NACK after it has sent a packet to the destination. For example, the destination will not send an ACK or a NACK if it does not receive the packet or if it receives the packet but is unable to determine the source for the packet. The latter situation can occur if there are errors in the source identication eld in the

490

GASS ET AL.

packet. Even if the destination does respond with an ACK or a NACK, the source may not receive the response. Because of such scenarios, an adaptive-transmission protocol must include a mechanism for adapting the transmission parameters in the event that the source makes a number of consecutive transmissions to a destination without receiving an ACK or a NACK. The mechanism employed in our adaptivetransmission protocol is as follows. If the source makes sufciently many consecutive transmissions to a destination without receiving an ACK or a NACK from that destination, the code rate is decreased if possible. If the lowest rate is already being used, the power level is increased if possible. The intent is that in the absence of a response from the destination, the source eventually transmits to the destination at the highest power level and lowest code rate. If there is no response to these transmissions, a higher level protocol will abandon this link and route packets around it until reliable communication is restored. Another need for a contingency measure arises if a number of consecutive packets fail to decode, even if the source receives a NACK for each of them. This can occur, for example, if the number of erasures exceeds n k in each of the error-control blocks. In the event that several consecutive receptions each have more than n k erasures, the adaptation procedure described above leads to the eventual use of the code of lowest rate, but it may not result in an increase in the transmitter power. The adaptive protocol deals with this problem in the same manner as described in the preceding paragraph. After a certain number of receptions have failed to decode, the rate of the code is reduced, if possible. If the code of lowest rate is being used, the power is increased if possible. One difference between the two mechanisms described above is in the terminal that initiates the procedure. Only the source can determine if an ACK or NACK is not received, so the mechanism to respond to missing ACKs or NACKs must be initiated by the source. The failure of packets to decode is determined by the destination, however, so the destination initiates the response to repeated decoding failures. 4. Integration with a reservation-based channel-access protocol Any reasonable channel-access protocol can be used with our adaptive-transmission protocol. For the numerical results presented in the next section, the channel-access protocol is based on a reservation technique in which there is an exchange of reservation messages prior to each packet transmission. In this section we describe the integration of the adaptive-transmission protocol and this type of channelaccess protocol. Adaptive transmission can benet from the use of a reservation-based channel-access protocol by exploiting the exchange of reservation messages to provide additional mechanisms for conveying side information and coordinating the selection of the transmission parameters. The messages exchanged in the channel-access protocol are a reservation-request packet and a conrmation packet.

If terminal 1 has a packet to send to terminal 2, it rst sends a request-to-send packet, referred to as an RTS. A eld in the RTS indicates that terminal 1 wishes to send a packet to terminal 2. If terminal 2 can accept the packet, it sends a clear-to-send packet, referred to as a CTS. The CTS gives terminal 1 permission to send the packet, and it may provide other information as well. Each terminal has two tables for each neighbor. In one table, the T-table, the terminal stores the transmission parameters to be used for transmissions of packets to each of its neighbors. In a second table, the R-table, the terminal stores the transmission parameters for packets to be received from each neighbor. When the source transmits a packet to the destination, the source determines the transmission parameters from its T-table for that destination. If the packet is received correctly by the destination and the ACK from the destination is received correctly by the source, the sources T-table for that destination agrees with the destinations R-table for that source. However, if the packet is not received correctly by the destination and the corresponding NACK is not received correctly by the source, these two tables may disagree. They may also disagree if the packet is received correctly by the destination but the corresponding ACK is not received correctly by the source. In either of these scenarios, it may be that one terminal may determine that one or more transmission parameters should be changed, so it updates the appropriate entries in its table. If the other terminal is unaware of the change, the two tables disagree until at least the next RTS-CTS exchange occurs. The method by which the source and destination make their determinations of the transmission parameters is described in the previous section. This description encompasses the scenarios in which packets fail to decode or there are missing ACKs or NACKs. Included in the current section is a discussion of how any disagreements between corresponding entries in a T-table and an R-table are resolved in those scenarios. The channel-access protocol and adaptive-transmission protocol interact in the process of establishing a reservation and selecting the transmission parameters. If the source has a packet to send to the destination, the source rst sends an RTS to the destination. Included in the RTS are the transmission parameters from the sources T-table for that destination. Upon receipt of the RTS the destination rst determines if it can accept the packet. For example, there can be no conicts between the proposed transmission and any prior reservations, and the destination must have space in its buffer to store the packet. If it agrees to accept the packet, the destination compares the transmission parameters in the RTS packet with the information in its R-table for the source. If one or more discrepancies exist, the destination selects the lower of the two code rates and the higher of the two power levels. Such a discrepancy can occur if an ACK or NACK from the destination is not received by the source or if there have been too many successive decoding failures. Once the destination selects the new transmission parameters, it updates its R-table with the new code and power information

ADAPTIVE-TRANSMISSION PROTOCOL FOR FREQUENCY-HOP NETWORKS

491

and sends the new parameters to the source as part of the CTS. Upon receipt of the CTS, the source updates its T-table with the code and power information contained in the CTS, and it then uses the new transmission parameters in the T-table to send the packet to the destination. Similarly, the destination uses the code information from its R-table to decode this packet. If the packet is received correctly, the destination determines the erasure and error counts and conveys the appropriate information to the source in an ACK. This information might be the error and erasure counts or the new transmission parameters that are derived from these counts. In either case, if the ACK is received correctly, the source learns which transmission parameters are proposed by the destination for the next packet to be sent to that destination. If the packet is not received correctly, the destination informs the source of this fact in a NACK. If the destination is able to determine the erasure and error counts, it conveys the appropriate information to the source in the NACK as described in the previous section. If the source receives this NACK, it updates its T-table accordingly and initiates the next transmission by sending the new transmission parameters to the destination in an RTS. For the numerical results presented in this paper, the destination decreases the code rate, if possible, if two successive packets fail to decode. If the lowest code rate is being used when two successive decoding failures occur, the destination increases the power level if possible. In each case, the destination updates its R-table accordingly. Missing ACKs or NACKs are handled as follows. If the source transmits a packet but receives neither an ACK or a NACK, the source decreases the code rate if possible. If the lowest code rate is already in use, the source increases the power level if possible. In each case, the source updates its T-table accordingly. 5. Performance results We evaluated the performance of the adaptive-transmission protocol by simulating a link in a frequency-hop spreadspectrum wireless communication network. Some of the results of our simulations are presented in this section. One link parameter is the excess path loss above some nominal value. This excess path loss might be used to model shadowing, for example. In our model the excess path loss on the link is either 0 dB (no excess path loss) or some nonzero (in dB) value. The numerical value for the non-zero excess path loss is not particularly important, and we use 3 dB in this paper for purposes of illustration only. The important parameter is the relative value of the non-zero excess path loss compared to the step size used to adapt the transmitter power. In our numerical results the step size is 1.5 dB. The codes employed in the adaptive-transmission protocol for the results presented in this paper are the (32, 24) and (32, 12) ReedSolomon codes, which have rates 3/4 and 3/8, respectively. Three transmitter power levels are available: P1 , P2 = P1 + 1.5 dB, and P3 = P1 + 3.0 dB. The

set S of transmission parameters available to the adaptivetransmission protocol consists of the two code rates and the three power levels. The adaptation parameters for our simulations are = 5, 1 = 107 , and 2 = 0.05. In practice the step-size for adapting the power should be selected according to the expected variations in the path loss on the communication link. The choice of 1.5 dB for the step size corresponds to an expected variation of 3 dB in the path loss. By providing two steps for increasing the transmitter power, we permit the adaptive-transmission protocol an alternative to increasing power by 3 dB in response to an increase in path loss. It has an option to increase the power by 1.5 dB and possibly reduce the code rate. The energy-to-noise-density ratio at the receiver in the simulation depends on the power in the transmission and the path loss of the link. For our simulation, the energy-to-noise density is specied as follows. If there is no excess path loss, the communication link in the simulation provides a received energy-to-noise-density ratio Es /N0 of 11 dB for a transmission at power level P1 , where Es is the energy in each binary symbol in the received packet. The values Es /N0 for other power levels and path-loss amounts can be determined from this value. There may be partial-band interference with onesided power spectral density 1 NI in a fraction of the band. The values of vary over the range from 0 to 0.4, and = 0 means there is no partial-band interference. If > 0 and there is no excess path loss, then the communication link in the simulation provides Es /N0 = 11 dB and Es /NI = 5 dB. An excess path loss of 3 dB on the link decreases each of these by 3 dB. The excess path loss and the parameter dene the state of the channel for any given packet transmission. The state is xed during the transmission of a packet. For a static channel the state never changes, but for a dynamic channel the state may change from one transmission to the next. The adaptive-transmission protocol is not given any information about the state of the channel. Its adaptation is based only on the erasure and error counts, as described in the previous section. Even for static channels, the adaptive-transmission protocol does not know the state of the channel, and its adaptation is based only on the counts of errors and erasures supplied by the receiving terminal. As a benchmark for performance comparisons, we consider a hypothetical adaptive-transmission protocol that has perfect knowledge of the channel state for the previous transmission. For each transmission this protocol uses the best transmission parameters from S for the state of the channel during the previous transmission. Thus, while this hypothetical adaptive-transmission protocol does not know the current state of the channel for a given transmission, it does know the previous state. Hence we refer to it as adaptive transmission with known previous state (KPS) or, more concisely, KPS adaptive transmission. If the channel state does not change from one transmission to the next, the KPS adaptive-transmission protocol uses the best transmission parameters from S. If the channel state changes, one transmission in the new state is required before the KPS protocol

492

GASS ET AL.

knows the new state and adapts its transmission parameters to match this state. For static channels the KPS adaptivetransmission protocol uses the best transmission parameters from S for each transmission, and therefore its performance is an upper bound on the achievable performance of any adaptive-transmission protocol. We also compare the performance of the adaptive-transmission protocol with a protocol that cannot change the transmission parameters. In order to ensure an acceptable error probability for each of the channel states, this protocol must use the low-rate code and the highest power level at all times. The choice of rate 3/8 and power level P3 is dictated by the worst channel state (i.e., 3 dB excess path loss and = 0.4). We refer to the resulting transmission protocol as the xed-transmission protocol. It represents a typical design philosophy for providing reliable communications in a system that cannot adapt to changes in the channel conditions. The performance measure that we employ to evaluate and compare different transmission protocols is the throughput efciency. In order to dene the throughput efciency, we introduce the concept of a valid information bit. A valid information bit is an information bit that is part of a packet that is received correctly. An information bit that is correct at the decoder output but is part of a packet in which one or more received words do not decode correctly is not a valid information bit. In other words, if the entire information packet is correct at the output of the decoder, each information bit in the packet is a valid information bit; otherwise, none of the information bits are valid. This is the most appropriate denition for packet transmission systems in which the entire information packet must be correct or else it has to be retransmitted. The throughput efciency is dened as the number of valid information bits at the decoder output of the destination per unit of transmitted energy. The transmitted energy for a given packet is the total energy for the initial transmission of a packet and any subsequent retransmissions of that packet. Acknowledgement packets and reservation packets are not included. The rst set of results is for a link whose channel state is xed but unknown to the terminals, except for the KPS protocol which has perfect knowledge of the state including the value of . There is no excess path loss, and partialband interference is present in a fraction of the band. The throughput efciencies obtained by the three different transmission protocols are compared in gure 1. The KPS protocol is able to use the parameters in S that provide the largest throughput efciency for the given channel state. As shown in gure 1, the adaptive-transmission protocol provides a much larger throughput efciency than the xed-transmission protocol for 0.35, and it provides a slightly larger throughput efciency for = 0.4. The higher throughput efciency for the adaptive-transmission protocol is a consequence of its ability to adapt to the unknown channel state by using the high-rate code and the lowest power level if is small. The adaptive-transmission protocol is able

Figure 1. Link throughput efciency for a static channel with partial-band interference.

to take advantage of favorable channel conditions if they exist. Since the xed-transmission protocol cannot adapt the code to match the value of , it is forced to use the lowrate code so that it can provide an acceptable error probability on a channel for which is large. Notice also that the adaptive-transmission protocol provides nearly as much throughput efciency as the KPS adaptive-transmission protocol for 0.3, which is the range of greatest interest for most applications. The success probability for a given transmission is dened to be the probability that each of the received words in the packet is decoded correctly. We examined the success probabilities corresponding to the throughput efciencies shown in gure 1, and we found that they are generally better for the adaptive-transmission protocol than for the KPS adaptive-transmission protocol. The success probabilities are about the same for these two protocols if 0 0.1 or 0.25 0.35. The adaptive-transmission protocol gives substantially better success probabilities than the KPS adaptive-transmission protocol if 0.15 0.2. In fact the success probabilities for the adaptive-transmission protocol are 0.9 or greater for = 0.15 and = 0.2, but at these two values of they are less than 0.72 for the KPS adaptive-transmission protocol. The success probabilities = 0.4 are 0.96 for the adaptive-transmission protocol and 0.91 for the KPS adaptive-transmission protocol. Thus, the larger throughput efciency achieved at = 0.4 by the KPS protocol is somewhat offset by the corresponding reduction in the success probability. The results illustrated in gure 2 are for a channel with a constant 3 dB excess path loss and partial-band interference in a fraction of the band. As shown in gure 2 the adaptive-transmission protocol has a higher throughput efciency than the xed-transmission protocol for all but the

ADAPTIVE-TRANSMISSION PROTOCOL FOR FREQUENCY-HOP NETWORKS

493

Figure 4. Two-state Markov chain for the time-varying excess path loss.

Figure 2. Link throughput efciency for a static channel with partial-band interference and 3 dB excess path loss.

Figure 3. Two-state Markov chain for intermittent partial-band interference.

Figure 5. Link throughput efciency for a dynamic channel with intermittent partial-band interference.

largest values of (i.e., 0.35), where the two give nearly the same performance. Notice that the throughput efciency for the adaptive-transmission protocol is nearly as large as the KPS adaptive-transmission protocol over the entire range 0 0.4. The second set of performance results are for a channel in which one or more parameters of the channel may vary from one transmission to the next. The two-state Markov model for time-varying partial-band interference, referred to as intermittent partial-band interference, is illustrated in gure 3. The two channel states correspond to a link with no partialband interference and a link with partial-band interference in a fraction of the band. The intermittent partial-band interference model is suitable for networks in which terminals may move through regions that have strong interference or for hostile jamming that employs directional antennas. At the time of each packet transmission the frequency-hop radios do not know the state of the channel nor do they know the value of . The adaptive-transmission protocol uses only the counts of errors and erasures as it adapts to the timevarying state and unknown value of . A two-state Markov model, illustrated in gure 4, is also used to test the adaptive protocols response to time-varying

excess path loss on the communication link. The two states of the Markov chain in gure 4 correspond to a channel with no excess path loss and a channel with 3 dB excess path loss. The throughput efciencies obtained for three transmission protocols operating over a dynamic channel with no excess path loss and intermittent partial-band interference are compared in gure 5. We see that for 0.4 the adaptivetransmission protocol has a much higher throughput efciency than the xed-transmission protocol. For 0.4 the adaptive-transmission protocol operates more efciently than the xed-transmission protocol by using the low power level and the high code rate when the channel switches to the state with no partial-band interference. When the channel switches to the state that has partial-band interference, the adaptive-transmission protocol adapts, if necessary, to a higher power or lower code rate. We see that for 0.4 the adaptive-transmission protocol provides throughput efciencies nearly as large as those obtained by the KPS adaptivetransmission protocol. The last set of performance results is for a dynamic channel in which both the partial-band interference and the excess path loss are time varying. Each of these is governed by its associated Markov chain (gures 3 and 4), and the

494

GASS ET AL.

mission, and these are easily determined in a frequency-hop receiver.

References
[1] M.B. Pursley and H.B. Russell, Network protocols for frequency-hop packet radios with decoder side information, IEEE Journal on Selected Areas in Communications 12 (May 1994) 612621. [2] M.B. Pursley and C.S. Wilkins, Adaptive-rate coding for frequencyhop wireless networks, in: Proceedings of the 1996 IEEE International Symp. Inform. Theory and Its Applications, Victoria, BC, Canada, Vol. 1 (September 1996) pp. 381384. [3] M.B. Pursley and C.S. Wilkins, An adaptive transmission method for frequency-hop radio networks, in: Proceedings of the 1997 Allerton Conf. on Commun., Control, and Computing, Monticello, IL (September 1997) pp. 713718. [4] M.B. Pursley and C.S. Wilkins, Adaptive-rate coding for frequencyhop communications over Rayleigh fading channels, IEEE Journal on Selected Areas in Communications 17 (July 1999) 12241232. [5] J.H. Gass, Jr., M.B. Pursley, H.B. Russell, R.J. Saulitis, C.S. Wilkins and J.S. Wysocarski, Adaptive transmission protocols for frequencyhop radio networks, in: 1998 IEEE Military Commun. Conf. Proceedings, Boston, MA, Vol. 2 (October 1998) pp. 282286. [6] M.B. Pursley, ReedSolomon codes in frequency-hop communications, in: ReedSolomon Codes and their Applications, eds. S.B. Wicker and V.K. Bhargava (IEEE Press, 1994) ch. 8, pp. 150174. [7] A.W. Lam and D.V. Sarwate, Comparison of two methods for generation of side information in frequency-hopping spread-spectrum multiple-access communications, in: Proceedings of the 21st Annual Conf. on Inform. Sciences and Systems, Johns Hopkins University (March 1987) pp. 426431. [8] A.J. Viterbi, A robust ratio-threshold technique to mitigate tone and partial band jamming in coded MFSK systems, in: 1982 IEEE Military Communications Conference Record (October 1982) pp. 22.4.15. [9] C.W. Baum and M.B. Pursley, Bayesian methods for erasure insertion in frequency-hop communications with partial-band interference, IEEE Transactions on Communications 40 (July 1992) 12311238.

Figure 6. Link throughput efciency for a dynamic channel with intermittent partial-band interference and time-varying path loss.

two Markov chains are statistically independent. As illustrated in gure 6 the adaptive-transmission protocol has a better throughput efciency than the xed-transmission protocol for each value of in the range 0 0.4. We see that for this channel the throughput efciencies for the adaptive-transmission protocol are approximately the same as those for the KPS adaptive-transmission protocol over the same range. 6. Conclusions We have demonstrated the improvement in energy efciency that can be achieved in a frequency-hop wireless network by employing an adaptive-transmission protocol that requires very little information from the previous transmission in order to select the parameters for the current transmission. We have shown that, even if the channel parameters are xed but unknown, the transmission parameters should be adapted to match the channel conditions in order to conserve energy and minimize transmission time. Performance results are also presented for dynamic channels whose parameters change according to rstorder Markov processes. The results presented in this paper demonstrate that for these dynamic channels the adaptivetransmission protocol achieves much greater energy efciency than can be obtained by transmitting with a xed power and code rate. Furthermore, for most of the ranges of channel parameters considered, the adaptive-transmission protocol gives about the same throughput efciency as a hypothetical protocol that has perfect knowledge of the channel state during the previous transmission. However, the adaptive-transmission protocol requires knowledge of only the counts of errors and erasures from the previous trans-

John H. Gass, Jr. received the B.S. degree (with honor) from the California Institute of Technology, Pasadena, CA, in 1991, the M.S. degree from the University of Illinois, Urbana-Champaign, in 1993, and the Ph.D. degree from Clemson University, Clemson, SC, in 1996, all in electrical engineering. He has been with the ITT Industries Aerospace/ Communications Division since May 1995 and is currently a staff engineer. His research interests are in the areas of spread-spectrum and multiple-access communications, communication over fading channels, and bandwidthefcient modulation. Dr. Gass is a member of the IEEE.

Michael B. Pursley received the B.S. degree (with highest distinction) and the M.S. degree, both in electrical engineering from Purdue University. He received the Ph.D. degree in electrical engineering from the University of Southern California. Dr. Pursleys industrial experience is primarily with the Space and Communications Group of the Hughes Aircraft Company during 19681974. He was a Hughes Doctoral Fellow and a Research Assistant in the Department of Electrical Engineering at the University of Southern California, and an Acting Assistant Professor in the System Science Department of the University of California, Los Angeles. From June 1974 through July 1993, he was with the Department of

ADAPTIVE-TRANSMISSION PROTOCOL FOR FREQUENCY-HOP NETWORKS

495

Electrical and Computer Engineering and the Coordinated Science Laboratory at the University of Illinois, Urbana, where he was promoted to the rank of Professor in 1980. Dr. Pursley is currently the Holcombe Professor of Electrical and Computer Engineering at Clemson University, Clemson, South Carolina. His research is in the general area of wireless communications with emphasis on spread-spectrum communications, applications of error-control coding, adaptive protocols for packet radio networks, and mobile wireless communication systems and networks. Dr. Pursley is a member of Phi Eta Sigma, Tau Beta Pi, and the Institute of Mathematical Statistics. He was elected Fellow of the Institute of Electrical and Electronics Engineers (IEEE) in 1982, and he is an honorary member of the Golden Key National Honor Society. He has held two three-year terms on the Board of Governors of the IEEE Information Theory Society, and he was elected president of that society in 1983. Dr. Pursley was a member of the Editorial Board of the Proceedings of the IEEE for the period 19841991. He is currently a member of Editorial Advisory Board for the International Journal of Wireless Information Networks, and he is a Senior Editor of the IEEE Journal on Selected Areas in Communications. He served as Technical Program Chairman for the 1979 IEEE International Symposium on Information Theory in Grignano, Italy, and he was a Cochairman for the 1995 IEEE International Symposium on Information Theory in Whistler, Canada. Dr. Pursley was awarded an IEEE Centennial Medal in 1984, and in 1996 he received the Ellersick Award of the IEEE Communications Society. In 1999 he received the IEEE Military Communications Conference Award for Technical Achievement. In 2000 he received an IEEE Millennium Medal and the Clemson University Alumni Research Award.

Harlan B. Russell received the B.S. degree (with high honors) in computer engineering in 1986 and the M.S. and Ph.D. degrees in electrical engineering in 1989 and 1993, respectively, all from the University of Illinois, Urbana-Champaign. From 1990 through 1999 he was with Techno-Sciences, Inc., and since 2000 he is an Assistant Professor of Electrical and Computer Engineering at Clemson University, Clemson, South Carolina. His research interests include adaptive protocols for packet radio networks and spread-spectrum communications. Dr. Russell is a member of the ACM and IEEE. E-mail: harlan.russell@ces.clemson.edu

Jeffrey S. Wysocarski received the B.S. degree (with great distinction) from Clarkson University, Potsdam, NY, in 1995, and the M.S. degree from Clemson University, Clemson, SC, in 1999, both in electrical engineering. He is currently pursuing the Ph.D. degree in electrical engineering at Clemson University, where he is also a Research Assistant in the Wireless Communications Program. His research interests are mobile wireless communication networks and adaptive protocols for packet radio networks. Mr. Wysocarski is a member of the IEEE.

Das könnte Ihnen auch gefallen