The online platform for Taylor & Francis Group content

Cookies Notification

This site uses cookies. By continuing to browse the site you are agreeing to our use of cookies. Find out more.
Advanced and citation search

Australian Journal of Electrical and Electronics Engineering

Cross-layer TCP/IP segmentation, re-routing and adaptive modulation techniques to exploit instantaneous BER variations on parallel subchannels
Original Articles

Cross-layer TCP/IP segmentation, re-routing and adaptive modulation techniques to exploit instantaneous BER variations on parallel subchannels

DOI:
10.1080/1448837X.2015.1093677
Erwin Anggadjajaa, Ian Vince McLoughlinb & Chiew Tong Lauc*

Abstract

This paper describes a cross-layer technique for TCP-enabled multichannel wireless communications systems such as Multiple-Input Multiple-Output (MIMO) links. The technique uses adaptive modulation (AM) and different packet re-routing strategies over physically decomposed sub-channels to exploit time-varying channel conditions. The paper will demonstrate that combining both AM and packet re-routing mechanisms can improve the overall performance of MIMO systems where sub-channels possess unbalanced or uneven time-varying error rate characteristics. A simulated wireless communications system is developed to evaluate packet re-routing techniques over four sub-channel connections made available to the link layer. It then considers the effect of cross-layer link modulation adaptation switching. The system is investigated primarily in terms of goodput for several channel arrangements and characteristics, for various data rates and mean bit error rates. Based on the investigation results, this paper will demonstrate that even simple re-routing and AM mechanisms may be beneficial for systems utilising multiple antennas, as long as these expose the underlying sub-channels and their different instantaneous error characteristics to the link layer.

Keywords

Related articles

View all related articles
 
This paper describes a cross-layer technique for TCP-enabled multichannel wireless communications systems such as Multiple-Input Multiple-Output (MIMO) links. The technique uses adaptive modulation (AM) and different packet re-routing strategies over physically decomposed sub-channels to exploit time-varying channel conditions. The paper will demonstrate that combining both AM and packet re-routing mechanisms can improve the overall performance of MIMO systems where sub-channels possess unbalanced or uneven time-varying error rate characteristics. A simulated wireless communications system is developed to evaluate packet re-routing techniques over four sub-channel connections made available to the link layer. It then considers the effect of cross-layer link modulation adaptation switching. The system is investigated primarily in terms of goodput for several channel arrangements and characteristics, for various data rates and mean bit error rates. Based on the investigation results, this paper will demonstrate that even simple re-routing and AM mechanisms may be beneficial for systems utilising multiple antennas, as long as these expose the underlying sub-channels and their different instantaneous error characteristics to the link layer.

Keywords

Spectrally efficient high data rate communications is the Holy Grail of future wireless communication systems, with Multiple-Input Multiple-Output (MIMO) technology often being suggested as the means to achieve such a goal. Without needing increased bandwidth or transmit power at the antennas, MIMO is capable of yielding higher data rate transmission (Paulraj et al. 200434. Paulraj, A. J., D. A. Gore, R. U. Nabar, and H. Bolcskei. 2004. “An Overview of MIMO Communications - A Key to Gigabit Wireless.” Proc. IEEE 92 (2): 198–218. ISSN 0018-9219. doi: 10.1109/JPROC.2003.821915.
[CrossRef], [Web of Science ®]
View all references
). Various early studies Foschini and Gans (199816. Foschini, G. J., and M. J. Gans. 1998. “On Limits of Wireless Communications in a Fading Environment When using Multiple Antennas.” Wireless Personal Communications 6: 311–335.
[CrossRef], [Web of Science ®]
View all references
); Telatar (199940. Telatar, E. 1999. “Capacity of Multi-antenna Gaussian Channels.” European Transactions on Telecommunications 10 (6): 585–595. ISSN 1541-8251. doi: 10.1002/ett.4460100604.
[CrossRef], [Web of Science ®]
View all references
) showed MIMO achieving remarkable spectral efficiency gains when compared to single-input single-output systems, assuming favourable channel conditions. Multiple antenna arrays at both ends of a MIMO link aim to exploit multipath propagation over the wireless channel, increasing throughput and improving reliability of transmission – an advantage that can be used to increase user data rate, as reported previously by Foschini and Gans (199816. Foschini, G. J., and M. J. Gans. 1998. “On Limits of Wireless Communications in a Fading Environment When using Multiple Antennas.” Wireless Personal Communications 6: 311–335.
[CrossRef], [Web of Science ®]
View all references
), Foschini (199617. Foschini, G. J. 1996. “Layered Space-time Architecture for Wireless Communication in a Fading Environment When Using Multi-element Antennas.” Bell Labs Technical Journal 1 (2): 41–59. ISSN 1536-1233. doi: 10.1109/TMC.2005.30.
[CrossRef]
View all references
).
Besides MIMO, another method of combating the effects of multi-path fading is to switch the transmission signal characteristics according to instantaneously changing channel conditions. For example, when the channel changes more slowly than a single packet transmission duration, the transmission characteristics of a subsequent packet can be adapted to suit the new conditions pertaining at the time of transmission. Or, adaption can be performed for a frame consisting of multiple packets. Adaptations can include adjusting coding rate/method, transmit power, or switching between multiple transmit antennas. In general, this technique of adapting parameters is known as link adaptation, whereas adjusting the signaling method is known as adaptive modulation (AM) (Camargo 20099. Camargo, A. 2009. Adaptive Modulation, Channel Coding and MIMO Schemes for Practical OFDM Systems. Germany: Shaker Verlag GmbH.

View all references
). Each of these is harnessing the ability to change code rate (or transmission power) to respond to channel conditions.
Some studies have shown that these techniques can improve transmission performance in the face of unbalanced bit error rates (BER) and spectral efficiency conditions (Goldsmith and Varaiya 199722. Goldsmith, A. J., and P. P. Varaiya. 1997. “Capacity of Fading Channels with Channel Side Information.” IEEE Transactions on Information Theory 43 (6): 1986–1992. ISSN 0018-9448. doi: 10.1109/18.641562.
[CrossRef], [Web of Science ®]
View all references
; Hayes 196824. Hayes, J. 1968. “Adaptive Feedback Communications.” IEEE Transactions on Communications and Technology 16 (1): 29–34. ISSN 0018-9332. doi: 10.1109/TCOM.1968.1089811.
[CrossRef], [Web of Science ®]
View all references
; Cavers 197210. Cavers, J. 1972. “Variable-rate Transmission for Rayleigh Fading Channels.” IEEE Transactions on Communications 20 (1): 15–22. doi: 10.1109/TCOM.1972.1091106.
[CrossRef], [Web of Science ®]
View all references
; Pottie 199535. Pottie, G. J. 1995. “System Design Choices in Personal Communications.” IEEE Personal Communications 2 (5): 50–67. doi: 10.1109/98.468362.
[CrossRef]
View all references
; Webb and Steele 199541. Webb, W. T., and R. Steele. 1995. “Variable Rate QAM for Mobile Radio.” IEEE Transactions on Communications 43 (7): 2223–2230. ISSN 0090-6778. doi: 10.1109/26.392965.
[CrossRef], [Web of Science ®]
View all references
; Alouini and Goldsmith 19971. Alouini, M. S., and A. Goldsmith. 1997. “Capacity of Nakagami Multipath Fading Channels.” IEEE 47th Vehicular Technology Conference; May; Vol. 1, 358–362. doi: 10.1109/VETEC.1997.596380.

View all references
, 19992. Alouini, M.-S., and A. Goldsmith. 1999. “Capacity of Rayleigh Fading Channels under Different Adaptive Transmission and Diversity-combining Techniques.” IEEE Transactions on Vehicular Technology 48 (4): 1165–1181. ISSN 0018-9545. doi: 10.1109/25.775366.
[CrossRef], [Web of Science ®]
View all references
). The results also span the flexibility of matching channel properties, i.e. coding, modulation, and signal and protocol parameters, to the conditions of the transmission link. Or in other words, the signal modulation and coding formats are altered to match instantaneous channel conditions, often measured in terms of receive-signal quality. Overall, the conclusion from these studies is that agile adjustments to signal and protocol parameters to suit varying radio link conditions can provide performance gains.
This paper implements various packet re-routing strategies for a MIMO link which employs the TCP protocol. This is then enhanced with cross-layer directed AM, and investigated in terms of the user-experienced goodput for various channel conditions.
We adopt the TCP/IP protocol implemented through the INET (http://inet.omnetpp.org) framework as part of an open source network simulation package using the OMNeT++ (http:/www.omnetpp.org) environment, to explore the model. The TCP/IP protocol will not be discussed in detail in this paper – however it is important to note that much research has been conducted on TCP and its (degraded) performance over wireless networks. Many possible improvements have been proposed. For example Bakre and Badrinath (19957. Bakre, A., and B. R. Badrinath. 1995. “I-TCP: Indirect TCP for Mobile Hosts.” In Proc. 15th Int. Conf. Distributed Computing Syst., Vancouver, June, 136–143.

View all references
), Chandran et al. (199811. Chandran, K., S. Ragbunathan, S. Venkatesan, and R. Prakash,. 1998. “A Feedback Based Scheme for Improving TCP Performance in ad-hoc Wireless Networks.” In Proc. 18th Int. Conf. Distributed Computing Systems, Amsterdam, May, 472–479. doi: 10.1109/ICDCS.1998.679778.

View all references
), Chen et al. (200412. Chen, C., H.-Q. Zhai, J.-F. Wang, and Y.-G. Fang. 2004. “TCP Performance Over Mobile ad hoc Networks.” Canadian Journal of Electrical and Computer Engineering 29 (1/2): 129–134. ISSN 0840-8688. doi: 10.1109/CJECE.2004.1425806.
[CrossRef], [Web of Science ®]
View all references
), Liu and Singh (200128. Liu, J., and S. Singh. 2001. “ATCP: TCP for Mobile ad hoc Networks.” IEEE Journal on Selected Areas in Communications 19 (7): 1300–1315. doi: 10.1109/49.932698.
[CrossRef], [Web of Science ®]
View all references
), including lower-layer enhancement (Anantharaman et al. 20044. Anantharaman, V., S.-J. Park, K. Sundaresan, and R. Sivakumar. 2004. “TCP Performance over Mobile ad hoc Networks: A Quantitative Study.” J. Wireless Commun. & Mobile Computing 4: 203–222. doi: 10.1002/wcm.172.
[CrossRef], [Web of Science ®]
View all references
; Fu et al. 200518. Fu, Z.-H., H.-Y. Luo, P. Zerfos, S.-W. Lu, L.-X. Zhang, and M. Gerla. 2005. “The Impact of Multihop Wireless Channel on TCP Performance.” IEEE Transactions on Mobile Computing 4 (2): 209–221. ISSN 1536-1233. doi: 10.1109/TMC.2005.30.
[CrossRef], [Web of Science ®]
View all references
), cross-layer design approaches, as well as new mechanisms which provide an underlying transport protocol similar to TCP (Sundaresan et al. 200539. Sundaresan, K., V. Anantharaman, H.-Y. Hsieh, and A. R. Sivakumar. 2005. “ATP: A Reliable Transport Protocol for ad hoc Networks.” IEEE Transactions on Mobile Computing 4 (6): 588–603. doi: 10.1109/TMC.2005.81.
[CrossRef], [Web of Science ®]
View all references
).
The system described in the paper utilises a payload-RDL (P-RDL) packet re-routing solution which has been published previously Anggadjaja, McLoughlin, and Premkumar (20136. Anggadjaja, E., I. V. Mcloughlin, and A. B. Premkumar. 2013. “TCP-based Multi Parallel Links Exploiting Packet Re-routing Mechanisms in Diverse Channel Condition.” In Wireless Conference (EW), Proceedings of the 2013 19th European, Guildford, 1–6.

View all references
). In summary, this adapts the handling of payload data over multiple sub-channels to adjust to local channel conditions, by requiring erroneous payloads to be retransmitted (R) over a different link (DL - hence ‘RDL’) as opposed to being retransmitted over the link that caused the packet error. ‘Payload’, in this context, refers to fragments of TCP packets, which are transmitted over air using a fast local-ARQ link strategy, employing a sliding window reconstruction mechanism. This was, in turn, a MIMO extension of previously published systems which developed a radio-link protocol (RLP) concept for multiple links (McLoughlin and Sirisena 201032. McLoughlin, I. V., and H. Sirisena. 2010. “TCP/IP Link Layer Error Mitigation for MIMO Wireless Links.” Journal of Telecommunication Systems 1–12. ISSN 1018-4864.

View all references
).
Any MIMO system relies upon multiple paths from transmitter to receiver at a physical level. These paths are typically processed within the physical layer so that it presents a single data pipe to the data link layer. In the cross-layer schemes described in this paper, these sub-channels implicit in the MIMO connection are maintained as raw transmission pipes to the data link layer, which is then able to exploit BER imbalances in those parallel data pipes by switching payload data between different links, in particular their treatment of retransmission following an error. The arrangement will be presented in detail in Section 3. It is evaluated specifically in terms of goodput improvement, and investigated in the light of diverse error rates between those links.
AM has been investigated widely by other authors as a solution to mitigate against channel propagation errors, with the main purpose being to match transmission rates to varying channel conditions (Alouini and Goldsmith 20003. Alouini, M.-S., and A. Goldsmith. 2000. “Adaptive Modulation Over Nakagami Fading Channels.” Wireless Personal Communications 13: 119–143. ISSN 0929-6212.
[CrossRef], [Web of Science ®]
View all references
; Goldsmith and Chua 199721. Goldsmith, A. J., and S.-G. Chua. 1997. “Variable-rate Variable-power MQAM for Fading Channels.” IEEE Transactions on Communications 45 (10): 1218–1230. ISSN 0090-6778. doi: 10.1109/26.634685
[CrossRef], [Web of Science ®]
View all references
, 199820. Goldsmith, A. J., and S.-G. Chua. 1998. “Adaptive Coded Modulation for Fading Channels.” IEEE Transactions on Communications 46 (5): 595–602. doi: 10.1109/26.668727.
[CrossRef], [Web of Science ®]
View all references
; Hole, Holm, and Oien 200025. Hole, K. J., H. Holm, and G. E. Oien. 2000. “Adaptive Multidimensional Coded Modulation Over Flat Fading Channels.” IEEE Journal on Selected Areas in Communications 18 (7): 1153–1158. doi: 10.1109/49.857915.
[CrossRef], [Web of Science ®]
View all references
; Chung and Goldsmith 200115. Chung, S.-T., and A. J. Goldsmith. 2001. “Degrees of Freedom in Adaptive Modulation: A Unified View.” IEEE Transactions on Communications 49 (9): 1561–1571. ISSN 0090-6778. doi: 10.1109/26.950343.
[CrossRef], [Web of Science ®]
View all references
; Choi and Hanzo 200314. Choi, B., and L. Hanzo. 2003. “Optimum Mode-switching-assisted Constant-power Single- and Multicarrier Adaptive Modulation.” IEEE Transactions on Vehicular Technology 52 (3): 536–560. doi: 10.1109/TVT.2003.810970.
[CrossRef], [Web of Science ®]
View all references
; Ong, Shikh-Bahaei, and Chambers 200833. Ong, L.-T., M. Shikh-Bahaei, and J. A. Chambers. 2008. “Variable Rate and Variable Power MQAM System based on Bayesian Bit Error Rate and Channel Estimation Techniques.” IEEE Transactions on Communications 56 (2): 177–182. ISSN 0090-6778. doi: 10.1109/TCOMM.2008.060142.
[CrossRef], [Web of Science ®]
View all references
), and even for the realistic scenario where only outdated channel information is available (Goeckel 199919. Goeckel, D. L. 1999. “Adaptive Coding for Time-varying Channels using Outdated Fading Estimates.” IEEE Transactions on Communications 47 (6): 844–855. ISSN 0090-6778. doi: 10.1109/26.771341.
[CrossRef], [Web of Science ®]
View all references
), or is corrupted by error (Prakash and McLoughlin 201136. Prakash, P. S., and I. V. McLoughlin. 2011. Analysis of Adaptive Modulation with Antenna Selection under Channel Prediction Errors. Calicut, India: In Proc. Int. Conf. Comms. and Signal Process.
[CrossRef]
View all references
). It has also been investigated as part of a MIMO orthogonal space–time block code (OSTBC) environment (Huang and Signell 200926. Huang, J.-L., and S. Signell. 2009. “On Performance of Adaptive Modulation in MIMO Systems Using Orthogonal Space Time Block Codes.” IEEE Transactions on Vehicular Technology 58 (8): 4238–4247. ISSN 0018-9545. doi: 10.1109/TVT.2009.2022475.
[CrossRef], [Web of Science ®]
View all references
). AM usually involves a switching mechanism rather than continuous adaptation, with its aim being to enhance the spectral efficiency of wireless systems while assuring QoS requirements. Much research has also been conducted regarding the implementation of AM. One prominent example in Liu et. al. Liu, Zhou, and Giannakis (200531. Liu, Q.-W., S.-L. Zhou, and G. B. Giannakis. 2005. “Queuing with Adaptive Modulation and Coding Over Wireless Links: Cross-layer Analysis and Design.” IEEE Transactions on Wireless Communications 4 (3): 1142–1153. ISSN 1536-1276. doi: 10.1109/TWC.2005.847005.
[CrossRef], [Web of Science ®]
View all references
) analysed the interaction of packet queuing at the data link with AM at the physical layer. A combination of AM with truncated ARQ protocol and limited retransmissions has been considered in Liu, Zhou, and Giannakis (2004a29. Liu, Q.-W., S.-L. Zhou, and G. B. Giannakis. 2004. “Cross-layer Combining of Adaptive Modulation and Coding with Truncated ARQ Over Wireless Links.” IEEE Transactions on Wireless Communications 3 (5): 1746–1755. ISSN 1536-1276. doi: 10.1109/TWC.2004.833474.
[CrossRef], [Web of Science ®]
View all references
), then improved by Gonzalez, Szczecinski, and Aissa (200423. Gonzalez, C., L. Szczecinski, and S. Aissa. 2004. “Throughput Maximization of ARQ Transmission Protocol Employing Adaptive Modulation and Coding.” In Telecommunications and Networking – ICT 2004. Vol. 3124 of Lecture Notes in Computer Science, edited by Jos de Souza, Petre Dini, and Pascal Lorenz, 33–44. Berlin/Heidelberg: Springer. ISBN 978-3-540-22571-3.
[CrossRef]
View all references
) which proposed a new approach to AM, referred to as Agressive AMC (A-AMC). Specific to the TCP layer, one novel concept named TCP “Adaptive-Selection” was proposed Caini, Firrincieli, and Lacamera (20088. Caini, C., R. Firrincieli, and D. Lacamera. March 2008. “The TCP ’Adaptive-selection’ Concept.” IEEE Systems Journal 2 (1): 83–89. doi: 10.1109/JSYST.2007.914744.
[CrossRef], [Web of Science ®]
View all references
) to extend AM into the transport layer. Further work Le, Hossain, and Le-Ngoc (200727. Le, L., E. Hossain, and T. Le-Ngoc. 2007. “Interaction between Radio Link Level Truncated ARQ, and TCP in Multi-rate Wireless Networks: A Cross-layer Performance Analysis.” IET Communications 1 (5): 821–830.ISSN 1751-8628. doi: 10.1049/iet-com:20060193.
[CrossRef], [Web of Science ®]
View all references
) showed a queuing model at the link layer through combining AM and ARQ for estimating the TCP layer throughput performance. Other research investigates the combination of TCP with AM (Liu, Zhou, and Giannakis 2004b30. Liu, Q.-W., S.-L. Zhou, and G. B. Giannakis. 2004. “TCP Performance in Wireless Access with Adaptive Modulation and Coding.” Proc. IEEE Int. Conf. Communications 7: 3989–3993. doi: 10.1109/ICC.2004.1313300.
[CrossRef]
View all references
; Rath, Karandikar, and Sharma 200837. Rath, H. K., A. Karandikar, and V. Sharma. May 2008. “Adaptive Modulation-based TCP-aware Uplink Scheduling in IEEE 802.16 Networks. In IEEE Int. Conf.” Communications 2008: 3230–3236. doi: 10.1109/ICC.2008.608.
[CrossRef]
View all references
; Selvam, Kathiravan, and Reshmi 200838. Selvam, A., K. Kathiravan, and R. Reshmi,. 2008. “A Cross-layer TCP Protocol with Adaptive Modulation for MANETs.” Int. Conf. Signal Processing, Communications and Networking, 2008; January, 428–433. doi:10.1109/ICSCN.2008.4447232.

View all references
).
This paper presents a pragmatic and efficient approach to combining AM with packet re-routing in a cross-layer arrangement. In particular it investigates the effect of adding discrete switched AM to a system implementing a P-RDL scenario (Anggadjaja, McLoughlin, and Premkumar 20136. Anggadjaja, E., I. V. Mcloughlin, and A. B. Premkumar. 2013. “TCP-based Multi Parallel Links Exploiting Packet Re-routing Mechanisms in Diverse Channel Condition.” In Wireless Conference (EW), Proceedings of the 2013 19th European, Guildford, 1–6.

View all references
). It assumes the simplest possible multi-channel communications system that presents multiple sub-channel transmission pipes to the data link layer. This is deliberately considered in the absence of any coding so that baseline results are presented (i.e. the performance of the re-routing and AM schemes are evaluated, rather than the strength of any particular individual coding mechanism). The method described is potentially compatible with many extensions to AM and TCP that have been proposed in the research literature, as mentioned previously. Similarly, the methods presented here can be combined with more advanced MIMO coding technologies – in practice the effect of these can be included in the simulation system model by simply reducing them to a set of new average BER figures for given modulation and data rate.
This section outlines the system configuration of the concepts applied in the models. Firstly assume two wired TCP nodes named S and R, connected over four physical or logical sub-channels that experience different error characteristics from time to time. We will name the subchannel links A, B, C and D.
We investigate packet switching mechanisms and retry scenarios for the four sub-channels, called the Quad-Parallel Independent Link Model (Quad-PILM), which is simply four independent sub-channels conveying segmented TCP/IP packets. Figure 1 illustrates the concept of the above-mentioned scenario, which shows four retry scenarios for four TCP data payloads that are each split into four segments before being transmitted. For each sub-figure, time advances from left to right. The subchannels from top to bottom are named A to D. The important concept is when considering the sequence of the transmitted segments and the identity of the subchannels. The details will be discussed in the next subsection.
Figure 1. Multiple-link arrangement showing how the erroneous segments from a single TCP/IP packet will be re-transmitted based upon different switching mechanisms. Erroneous transmission is shown with an ‘’.

3.1. Segmented-window switching over different links

The left side of Figure 1 (subplots (i) and (iii)) shows two Segmented-Window switching (SWS) mechanisms. In these, by default, the first segment (a) will be sent over Link A. This will be followed by the subsequent segment (b) on Link B, and so on before switching back again in a round-robin fashion until all segments have been successfully transmitted. In all of the subplots, segments from subsequent TCP/IP packets are not shown on the diagram for reasons of clarity, however they will normally be fitted into all of the available space in the diagram.
Once any segment experiences an error (shown in Figure 1 by an ‘’) then a retry will be attempted. In subplot (i), the system will cause the erroneous segment, which employed SWS mechanism, to be retransmitted over a different link (RDL) to the one which experienced the error (for the rest of paper this would be mentioned as S-RDL). Similarly, subplot (iii) shows retransmission (R) over a randomly chosen link (RL, and hence ‘S-RRL’), which may occasionally cause the segment to be transmitted over the same link.
In the S-RRL scenario, in practice the system will retransmit over whichever link next becomes available since this is randomised in the simulations. In this way, erroneous segments are able to utilise whatever link may be available.

3.2. Payload switching over different links

The payload (PLD) switching scheme is shown in Figure 1 (right hand subplots). By contrast, this uses higher layer knowledge to force each segmented window from one particular TCP payload to be conveyed over the same link as all others derived from the same TCP payload. This has been shown to be advantageous in certain circumstances when the BER of each link differs (McLoughlin and Sirisena 201032. McLoughlin, I. V., and H. Sirisena. 2010. “TCP/IP Link Layer Error Mitigation for MIMO Wireless Links.” Journal of Telecommunication Systems 1–12. ISSN 1018-4864.

View all references
). In this case, segmented windows from the next TCP payload will be switched to transmit over the other link.
Now considering retries also, the trend is similar to the SWS case: P-RDL re-sends erroneous segments to another one of three remaining links, while P-RRL sends over a random link. Note that, for the random link scenarios, since this always means that when a packet is sent twice (i.e. resent once), it utilises both the better and the worse link, the errors become more evenly spread, and then the overall error rate will better match the average of the BERs exhibited by each link.
Note also that Figure 1 is simplified. In practice, as mentioned, with many TCP packets being transmitted there would not be any idle slots. When simulating these systems, equal bandwidth links are specified for all arrangements, but the parallel links will differ slightly in their instantaneous BERs.

3.3. Segmentation and reassembly of TCP data

Two wireless transmitters named X and Y are ‘inserted’ between the previously described nodes S and R (Figure 2) which are designed to be invisible to TCP (Anggadjaja and Mcloughlin 20105. Anggadjaja, E., and I. V.Mcloughlin. 2010. “Point-to-point OMNeT++ Based Simulation of Reliable Transmission Using Realistic Segmentation and Reassembly with Error Control.” 2nd International Confrence on Advances in Computing, Control and Telecommunication Technologies (ACT), Jakarta, December, 125–128. doi: 10.1109/ACT.2010.25.

View all references
). All of the normal TCP/IP aspects such as connection establishment, termination, flow, error control mechanisms, and also the concepts of encapsulation and de-encapsulation from the layers above and below are supported transparently by X and Y.
Within the simulation, a large data block to be transferred from node S to node R is first divided into a sequence of TCP packets at node S in the usual way. If this sequence is transmitted and received without error by node R, it will be reassembled into the received TCP data block. During the transmission process, TCP packets from source node S travel to node X where they are fragmented into smaller segments for sending over the wireless link to node Y. A one-byte sequence number is appended to each segment, before it is transmitted over air from node X, to node Y, where it is to be reassembled. This segmentation and reassembly (SAR) arrangement is shown in Figure 3.
Figure 3. Illustration of a TCP/IP and SAR mechanisms.

During operation, source node S transmits TCP/IP data to receiver node R through intermediate nodes X and Y as mentioned. X and Y have the responsibility for ‘hiding’ the potentially high BER wireless link X–Y from nodes S and R. Each of the parallel sub-channels A to D forms an independent digital link. These links will each experience unique but slightly differing error characteristics.1 It is these different error characteristics that the techniques of this paper hope to exploit.

4.1. Instantaneous channel differences

To explore the techniques described, a simulation is constructed in which we artificially adjust the BER experienced by each channel individually, but constrain mean BER to a specified level. In contrast to real world systems which would naturally experience instantaneously changing BER, our simulations fix unequal BERs for an entire simulation run, repeating this for a range of degrees of inequalities – all the while maintaining overall mean BER. Important parameters in this arrangement are thus the BER mean and variance. In practical systems, the BER inequality would change over time, rapidly in some situations, slowly in others. A sequence of frames would each experience different error rates, however they would likewise be described by their BER mean and variance, for a given mean SNR. In fact these inputs to our simulation (i.e. mean BER and BER difference) are commonly and routinely derived in research papers which propose novel wireless coding methods, or investigate existing ones. Likewise, real implemented systems would be characterised using similar data.
Therefore, this model assumes that links experience unequal error rates, i.e. where one channel is slightly worse than others, but will always ensure that the overall mean BER is as specified for the given SNR. Also that the total number of errors during data transmission is consistent with this average. The BER variance is explored in terms of a channel difference: namely being expressed as the difference between mean and instantaneous BER over a given transmission frame.
In general, for B bits transmitted over each of L channels with a mean BER denoted by , assuming the then, the number of errors E experienced should tend towards:
(1)
We now introduce parameter n to determine the channel difference distribution (i.e. the instantaneous BER difference between channels). This is used to set one sub-channel (named the ‘controlled’ channel) to have a different BER to others. For example, given sub-channels, the controlled BER of the single different controlled channel will be:
(2)
while the remaining links have the following BER:
(3)
The imbalance is therefore described as a balancing comparison, i.e. one sub-channel is controlled in terms of its BER, while the remaining errors needed to achieve an average BER target are spread equally among the remaining three links. By doing this, we can maintain overall BER for all channel difference scenarios. So that, since all links are transmitting continuously, for a four-channel system, the total errors are given as:
(4)
which is consistent with Eqn. 1
Table 1. Quad-link error distribution for different channel difference values n, when =1000.
Using (2) and (3) and taking an example of , then we would have the BER distribution for four channels accordingly as stated in Table 1. In all scenarios (i.e. the columns in Table 1), it should be clear that the system average BER (and total number of errors per unit time) remains constant.

4.2. Evaluation method

In this paper, goodput is used to evaluate the effects of adjusting and changing parameters in the various scenarios tested. In the context of this paper, it only considers successfully received bytes of the payload at the receiver – excluding the protocol overhead and erroneous frames.
Consider n as the number of correctly transmitted bytes over total time duration of N bytes of data, then goodput GP is a measure of the amount of useful data received per unit time:
(5)
Since the system is packet-based and without any form of forward-error correction (FEC), a single corrupt byte in one data packet results in the entire packet being discarded (an argument for smaller packets (Anggadjaja, McLoughlin, and Premkumar 20136. Anggadjaja, E., I. V. Mcloughlin, and A. B. Premkumar. 2013. “TCP-based Multi Parallel Links Exploiting Packet Re-routing Mechanisms in Diverse Channel Condition.” In Wireless Conference (EW), Proceedings of the 2013 19th European, Guildford, 1–6.

View all references
)).
Figure 4. n segment transmissions each contain N bytes of data. Segments includes a one byte sequence number. Errors cause some segments to be re-transmitted, maybe more than once. transmissions will be necessary in total to convey all data.

In the other words, if any constituent segment of a TCP/IP frame is missing after all retries have been exhausted, then the entire TCP/IP frame will be discarded. This happens even if all other segments have been received correctly; none of the good data (from the partial TCP frame) will be counted towards goodput (refer to Figure 4).

4.3. Analysis and discussion

Simulations were conducted to highlight the difference in performance due to changing switching methods in Quad-PILM for poor BER conditions. Figure 5 plots goodput against the degree of channel difference, and clearly shows that for Quad-PILM, switching methods S-RDL and S-RRL both finally outperform the original SWS method as the degree of difference exceeds 1.0. Note that channel difference needs to be explored both below and above 1.0 because it is not symmetrical since only a single channel is being changed with respect to the remaining three.
Between the improvements achieved by the switching methods, it is noticeable that when the system delivers the retry frames into a random link (RRL), it performs slightly worse than when retries are delivered into any three other links apart from the erroneous one (RDL). This can be explained as being due to the latter not re-transmitting the frame into the current failed link; while in RRL, there is always a chance the frame is delivered again over the failed link.
Figure 5. SWS-switching in a Quad-PILM scenario.

Figure 6. PLD-switching in a Quad-PILM scenario.

Similar trends also apply to the payload-switching methods (see Figure 6). When the channel difference value exceeds 1.0, the original PLD performs worst compared to the improved PLD methods (namely P-RRL and P-RDL). The RDL scenario, as expected, will prevent the re-transmitting frame from being delivered into the same susceptible link, and thus will slightly surpass the RRL scenario.
Figure 7. Quad-PILM without AM for = 4 and 6 times respectively.

Figure 8. Fixed modulation versus AM for AM-LV with various settings for .

Figure 9. Fixed modulation versus AM-LV scenario, with various settings for .

Figure 10. Effects of for fixed modulation vs. AM-LV.

Based upon these, we conclude that the PLD method with the RDL retry scenario is the best approach for the given conditions (which is consistent with (Anggadjaja, McLoughlin, and Premkumar 20136. Anggadjaja, E., I. V. Mcloughlin, and A. B. Premkumar. 2013. “TCP-based Multi Parallel Links Exploiting Packet Re-routing Mechanisms in Diverse Channel Condition.” In Wireless Conference (EW), Proceedings of the 2013 19th European, Guildford, 1–6.

View all references
)). PLD with RDL will thus be adopted for the subsequent simulations reported in this paper.
To improve performance of the system over P-RDL alone, we propose equipping the basic P-RDL re-routing system with the capability to perform cross layer-directed AM. The four sub-channel simulation (Quad-PILM) (Anggadjaja, McLoughlin, and Premkumar 20136. Anggadjaja, E., I. V. Mcloughlin, and A. B. Premkumar. 2013. “TCP-based Multi Parallel Links Exploiting Packet Re-routing Mechanisms in Diverse Channel Condition.” In Wireless Conference (EW), Proceedings of the 2013 19th European, Guildford, 1–6.

View all references
) is maintained as the evaluation baseline, although the AM technique is of course be applicable to a wide variety of other wireless systems.
In this system, the current BER values are determined over either each parallel sub-channel, or over the link as a whole and this information is used to direct an adjustment to the QAM modulation depth. The exact values of BER obtained with respect to SNR are obviously system dependent. For simulation purposes, we will choose standard BER values from a benchmark system as reported by Cho and Yoon Cho and Yoon (200213. Cho, K., and D. Yoon. 2002. “On the General BER Expression of One- and Two-dimensional Amplitude Modulations.” IEEE Transactions on Communications 50 (7): 1074–1080. doi: 10.1109/TCOMM.2002.800818.
[CrossRef], [Web of Science ®]
View all references
), which are reproduced in Table 2. These inputs to our simulation could very easily be changed to investigate the effects of applying the method to other wireless systems.
The table is used for combining and analysing the different BER values over a fixed link in which QAM modulation is used. Also, it also provides the ability to adjust the QAM level.
Table 2. BER value for different QAM modulations for 15dB SNR. Data extracted from Cho and Yoon Cho and Yoon (200213. Cho, K., and D. Yoon. 2002. “On the General BER Expression of One- and Two-dimensional Amplitude Modulations.” IEEE Transactions on Communications 50 (7): 1074–1080. doi: 10.1109/TCOMM.2002.800818.
[CrossRef], [Web of Science ®]
View all references
).
The scenario in this work is assuming that a channel that experiences persistent errors (most likely it is the worst channel) to adjust its modulation (i.e., halving or doubling the data rate). When the adjustment is made, the BER experienced by that link will naturally change accordingly as stated in Table II, as will the data rate over the link. Channels will therefore independently adapt their modulation format.2
It should be noted at this point that the simulations reported in this paper begin with the highest QAM level, and force only downward steps in adjustment to simplify the subsequent analysis of results.

5.1. Quad-PILM for selected modulations

Initially, for benchmarking purposes, we derive a normal result for each QAM modulation, unswitched (i.e. without any AM adjustments being made). These yield performance bounds. For each simulation run, we plot the goodput, and repeat for two different retry values ( from Anggadjaja, McLoughlin, and Premkumar (20136. Anggadjaja, E., I. V. Mcloughlin, and A. B. Premkumar. 2013. “TCP-based Multi Parallel Links Exploiting Packet Re-routing Mechanisms in Diverse Channel Condition.” In Wireless Conference (EW), Proceedings of the 2013 19th European, Guildford, 1–6.

View all references
)). TCP Reno is employed with the channel data rate and segmented frame sizes set to 256 kbps and 128 bytes respectively. For each simulation run, a 30 MB file is transmitted with maximum TCP-payload retransmission set to 12 times.
Figure plots results, with the horizontal axis exploring the effect of channel difference as specified earlier. Graphs shows an unsurprising variation in achieved goodput for the differing modulation conditions. It is notable that a bigger retry value of 6 times results in more segments being re-sent on error, and thus there is a higher probability that they will be correctly received. This results in improved goodput compared to the 4 times retry limit. Naturally, P-RDL is successful in improving goodput through exploiting the degree of channel difference, and we can observe this effect as a gradual improvement from left to right. Remember that AM is not operating in this system – the results are plotted purely as a baseline which explores the different modulation settings that are available.

6.1. AM-scenario

Proceeding to explore AM, we will introduce a simple parameter called NAK, which is used to keep track of the number of retries that have occurred at the sender (or from the receiver’s perspective, it counts the total number of corrupted or missing segments). The intention is that NAK should increase during data transmission, meaning that the communication link is poor. This will be used (specifically, when it reaches a cumulative threshold over a fixed period, i.e. ) to trigger the system to adjust modulation. In this simulation, once triggered, the modulation rate will be reduced, and consequently reduce the achieved data rate by one step, while improving the experienced mean BER.
The adjustment process can be explained as follows. When one link fails or performs consistently badly, NAK will increment until it reaches or exceeds the threshold , at which point the adjustment will be made according to the settings in Table 2 by reducing one modulation level. This operation will also reset the accumulated NAK counter for the link. If is not reached, the system will simply continue transmitting as before. The normal retry mechanism will remain in force, resending the failed segments up to the maximum allowed number of retries.
Although at face value this scheme appears reasonable, and does improve performance when simulated, it includes no way to determine which link to adjust. Although we control the BER of one link in the experiment compared to the rest (through the channel difference parameter n), the system itself does not reliably know which link is degraded. This is complicated by the fact that, as modulation depth (and hence bandwidths) change, a different number of packets may be transmitted over each link.
This highlights the importance of system-level design considerations, and prompts a different method of directing the AM to be determined, as described in the next subsection.

6.2. AM using individual link value parameters

Another parameter, named link value (), is introduced to track the performance of each overall link individually in terms of packet error-rate (for ). These values are incremented for each successfully transmitted payload, and decremented for each un-successful one over each individual link. As before, the aim for this is to derive a switching rule. Whenever a switch occurs, all LV parameters are re-initialised.
A per-channel value, operating as specified in the previous subsection, is again used as the determinator of when to adjust modulation (i.e. when it exceeds threshold ), but now it will no longer control which link to adjust since that is the role of the counters. The link with the lowest LV value is the one switched to a lower modulation. This system now performs longer term global performance tracking, but with a shorter-term local per-channel indication of performance. In this scheme, which we name AM-LV, whilst occasionally the incorrect link will be adjusted, it is a simple, portable and pragmatic method of tracking adjustments. An alternative will be presented in the following subsection.
Simulations were performed for different values of , and compared against the non-AM benchmarks. The results are plotted in Figure 8. Although, it was found that with , no adjustment occurred throughout the duration of each simulation, when setting , adjustments were made, showing improvement to the results curve compared to the baseline.
The bottom three lines of Figure 8 appear to show that two of the AM scenarios yield similar results to the fixed modulation benchmark. To attempt to explain this result, remember that the maximum number of retries allowed, , and , were set to 4 and 3 respectively. Since applies for each link, the system will have to reach a global value which, for extreme cases, is as high as 12 retries, compared to a lower maximum retry value, i.e. 4. Thus it is very possible that the system will simply discard the segments (and automatically the payload) once a maximum retry value is reached, before any AM adjustment is made. The effect on goodput of discarding an entire TCP payload is so extreme, that this masks the smaller and slower gains due to performing AM.
However by setting , the AM adjustment is performed earlier (on average), allowing a significant improvement in goodput to be realised. The AM-scenario is actually working as a very early ‘filter’ to force the system to perform the adjustment before significant link retry-and-timeout occurs. As a consequence, when any of the links experience error, the adjustment happens as soon as its NAK value reaches 2, increasing the overall chance that segments arrive safely at the receiver.
Although an improvement in goodput is achieved in the simulation, we consider to be too premature to force a correct adjustment (it is acting upon insufficient information to make a correct choice in many cases, since the probability of selecting the incorrect channel for modulation adjustment will be high). We thus proposed altering the system to counter this problem as will be shown in the next subsection. However the results presented here stand to demonstrate the potential of performing AM adjustment using a cross-layer methodology based effectively on PER: the method has been shown to be capable of significant performance improvement under certain operating circumstances.

6.3. Modified AM-scenario based on link error value

Unfortunately the size of packets in the RLP-like simulation are designed to be small so as to reduce the PER experienced for given BER. Since the AM adjustment is based on counting the total NAKs at each link, which is effectively a measure of PER, this becomes an insensitive method. Having the threshold value of 2 and only small BER differences between channels, it is very easy for an incorrect channel to be flagged as the worst-performing one, and to receive an incorrect modulation adjustment.
We therefore develop another scenario which counts the global NAKs at the transmission node itself – we will name this AM-LV. In this system, whenever any link experiences an error, a global NAK counter (instead of local for each link in the previous scenario) will be incremented. Similarly, for each channel will be updated as before. Operation continues unchanged until the base NAK value reaches the threshold . At this point, the system will decide which link has the worst value by comparing values, and will subsequently perform the adjustment on the link having the lowest LV. If two links have the same LV, the system simply adjusts whichever of the two links performed worst last time. For a quad link system, this ensures that there are now approximately four times as many NAKs before reaching a threshold, thus making the cutoff decision more sensitive.
Again, we compare the scenario to the non-AM scheme, with the results shown in Figure 9. First, we see a significant goodput improvement for all values compared to the fixed modulation benchmark. In general, we can see that AM-LV works better under the tested conditions (i.e. for various values) than AM-LV ; with the adjustment scheme being well utilised. For moderate channel differences (between 0.9 and 1.2), since all the channels have similar BER, the probability of adjustment is very low, and thus the goodput improvement is not great.
However we found that for more extreme channel differences, i.e. when the channel difference distribution is below 0.9 or larger than 1.2, goodput improves significantly compared to moderate channel differences. This indicates that the updated AM scenario (AM-LV) works perfectly, especially in seriously unbalanced link cases. When channel difference is large, the probability of adjustment increases – and this also increases the possibility of the frames being communicated without error. The graph shows a step change in goodput performance based solely upon a thresholding effect on the experienced packet error rate.
Looking closely at the different values on the plot, we see evidence that for large thresholds, the system performs well using an AM-LV scheme. Bigger values obviously give the system more opportunity to perform more error recovery prior to adjustments being made. Clearly, we can see this on the most extreme channel distribution (top lines), shown in Figure 10 which plots goodput against for several channel difference values.
In this paper, cross-layer packet handling and modulation adjustment mechanisms have been explored for generic TCP-enabled multiple channel communications systems through OMNeT++ simulation. In the simulations, a MIMO-like model named Quad-PILM combines packet re-routing with a simple switched AM scheme. The TCP protocol model is adopted in the simulations, with user-experienced goodput chosen as the basis for assessing performance. In the P-RDL model used in the simulations, packet routing and modulation adjustment are based on higher layer BER (and/or PER) statistics, rather than lower layer information such as SNR, which is more common. Several variants of a basic AM adjustment model were constructed to explore goodput. The results show that tracking individual link performance to determine which link to adjust modulation for, along with a global trigger for performing the adjustment, works well for the tested channel conditions. In particular, significant goodput improvement was exhibited for quite extreme link differences, but less for moderate channel differences. This indicates that the system is indeed able to exploit channel difference in order to improve performance. Results validate the proposed AM model and reveal that even simple switching heuristics can be beneficial when implementing TCP-based data communications over a multiple antenna system deployment in which sub-channel transmission pipes are made available to the data link layer. The systems discussed in this paper are shown to be quite capable of exploiting the natural time-varying differences in the instantaneous BER over different sub-channels, as long as these vary slowly enough so that the BER statistics remain pseudo-stationary over the period of analysis.

Notes

No potential conflict of interest was reported by the author(s).
1 Data streams conveyed by the physical layer are usually characterised by their mean BER. In reality the BER of each sub-channel will vary away from this mean in real deployed wireless systems (McLoughlin and Sirisena 201032. McLoughlin, I. V., and H. Sirisena. 2010. “TCP/IP Link Layer Error Mitigation for MIMO Wireless Links.” Journal of Telecommunication Systems 1–12. ISSN 1018-4864.

View all references
): MIMO coding naturally exploits this variability. Similarly the techniques described in this paper also exploit the variability where it remains stable for one transmission frame or longer – which would be the norm in the majority of deployed systems that do not continually operate at or beyond their worse-case channel conditions.
2 At this point it should be observed that, apart from the normal time-varying instantaneous BER imbalance between sub-channels in real-world MIMO systems, there can sometimes be long-term or static imbalances, which may be caused by sub-optimal placement of one or more antennas, by obstructions, or by the tolerances of electronic components

References

  • 1. Alouini, M. S., and A. Goldsmith. 1997. “Capacity of Nakagami Multipath Fading Channels.” IEEE 47th Vehicular Technology Conference; May; Vol. 1, 358–362. doi: 10.1109/VETEC.1997.596380.
  • 2. Alouini, M.-S., and A. Goldsmith. 1999. “Capacity of Rayleigh Fading Channels under Different Adaptive Transmission and Diversity-combining Techniques.” IEEE Transactions on Vehicular Technology 48 (4): 1165–1181. ISSN 0018-9545. doi: 10.1109/25.775366. [CrossRef], [Web of Science ®]
  • 3. Alouini, M.-S., and A. Goldsmith. 2000. “Adaptive Modulation Over Nakagami Fading Channels.” Wireless Personal Communications 13: 119–143. ISSN 0929-6212. [CrossRef], [Web of Science ®]
  • 4. Anantharaman, V., S.-J. Park, K. Sundaresan, and R. Sivakumar. 2004. “TCP Performance over Mobile ad hoc Networks: A Quantitative Study.” J. Wireless Commun. & Mobile Computing 4: 203–222. doi: 10.1002/wcm.172. [CrossRef], [Web of Science ®]
  • 5. Anggadjaja, E., and I. V.Mcloughlin. 2010. “Point-to-point OMNeT++ Based Simulation of Reliable Transmission Using Realistic Segmentation and Reassembly with Error Control.” 2nd International Confrence on Advances in Computing, Control and Telecommunication Technologies (ACT), Jakarta, December, 125–128. doi: 10.1109/ACT.2010.25.
  • 6. Anggadjaja, E., I. V. Mcloughlin, and A. B. Premkumar. 2013. “TCP-based Multi Parallel Links Exploiting Packet Re-routing Mechanisms in Diverse Channel Condition.” In Wireless Conference (EW), Proceedings of the 2013 19th European, Guildford, 1–6.
  • 7. Bakre, A., and B. R. Badrinath. 1995. “I-TCP: Indirect TCP for Mobile Hosts.” In Proc. 15th Int. Conf. Distributed Computing Syst., Vancouver, June, 136–143.
  • 8. Caini, C., R. Firrincieli, and D. Lacamera. March 2008. “The TCP ’Adaptive-selection’ Concept.” IEEE Systems Journal 2 (1): 83–89. doi: 10.1109/JSYST.2007.914744. [CrossRef], [Web of Science ®]
  • 9. Camargo, A. 2009. Adaptive Modulation, Channel Coding and MIMO Schemes for Practical OFDM Systems. Germany: Shaker Verlag GmbH.
  • 10. Cavers, J. 1972. “Variable-rate Transmission for Rayleigh Fading Channels.” IEEE Transactions on Communications 20 (1): 15–22. doi: 10.1109/TCOM.1972.1091106. [CrossRef], [Web of Science ®]
  • 11. Chandran, K., S. Ragbunathan, S. Venkatesan, and R. Prakash,. 1998. “A Feedback Based Scheme for Improving TCP Performance in ad-hoc Wireless Networks.” In Proc. 18th Int. Conf. Distributed Computing Systems, Amsterdam, May, 472–479. doi: 10.1109/ICDCS.1998.679778.
  • 12. Chen, C., H.-Q. Zhai, J.-F. Wang, and Y.-G. Fang. 2004. “TCP Performance Over Mobile ad hoc Networks.” Canadian Journal of Electrical and Computer Engineering 29 (1/2): 129–134. ISSN 0840-8688. doi: 10.1109/CJECE.2004.1425806. [CrossRef], [Web of Science ®]
  • 13. Cho, K., and D. Yoon. 2002. “On the General BER Expression of One- and Two-dimensional Amplitude Modulations.” IEEE Transactions on Communications 50 (7): 1074–1080. doi: 10.1109/TCOMM.2002.800818. [CrossRef], [Web of Science ®]
  • 14. Choi, B., and L. Hanzo. 2003. “Optimum Mode-switching-assisted Constant-power Single- and Multicarrier Adaptive Modulation.” IEEE Transactions on Vehicular Technology 52 (3): 536–560. doi: 10.1109/TVT.2003.810970. [CrossRef], [Web of Science ®]
  • 15. Chung, S.-T., and A. J. Goldsmith. 2001. “Degrees of Freedom in Adaptive Modulation: A Unified View.” IEEE Transactions on Communications 49 (9): 1561–1571. ISSN 0090-6778. doi: 10.1109/26.950343. [CrossRef], [Web of Science ®]
  • 16. Foschini, G. J., and M. J. Gans. 1998. “On Limits of Wireless Communications in a Fading Environment When using Multiple Antennas.” Wireless Personal Communications 6: 311–335. [CrossRef], [Web of Science ®]
  • 17. Foschini, G. J. 1996. “Layered Space-time Architecture for Wireless Communication in a Fading Environment When Using Multi-element Antennas.” Bell Labs Technical Journal 1 (2): 41–59. ISSN 1536-1233. doi: 10.1109/TMC.2005.30. [CrossRef]
  • 18. Fu, Z.-H., H.-Y. Luo, P. Zerfos, S.-W. Lu, L.-X. Zhang, and M. Gerla. 2005. “The Impact of Multihop Wireless Channel on TCP Performance.” IEEE Transactions on Mobile Computing 4 (2): 209–221. ISSN 1536-1233. doi: 10.1109/TMC.2005.30. [CrossRef], [Web of Science ®]
  • 19. Goeckel, D. L. 1999. “Adaptive Coding for Time-varying Channels using Outdated Fading Estimates.” IEEE Transactions on Communications 47 (6): 844–855. ISSN 0090-6778. doi: 10.1109/26.771341. [CrossRef], [Web of Science ®]
  • 20. Goldsmith, A. J., and S.-G. Chua. 1998. “Adaptive Coded Modulation for Fading Channels.” IEEE Transactions on Communications 46 (5): 595–602. doi: 10.1109/26.668727. [CrossRef], [Web of Science ®]
  • 21. Goldsmith, A. J., and S.-G. Chua. 1997. “Variable-rate Variable-power MQAM for Fading Channels.” IEEE Transactions on Communications 45 (10): 1218–1230. ISSN 0090-6778. doi: 10.1109/26.634685 [CrossRef], [Web of Science ®]
  • 22. Goldsmith, A. J., and P. P. Varaiya. 1997. “Capacity of Fading Channels with Channel Side Information.” IEEE Transactions on Information Theory 43 (6): 1986–1992. ISSN 0018-9448. doi: 10.1109/18.641562. [CrossRef], [Web of Science ®]
  • 23. Gonzalez, C., L. Szczecinski, and S. Aissa. 2004. “Throughput Maximization of ARQ Transmission Protocol Employing Adaptive Modulation and Coding.” In Telecommunications and Networking – ICT 2004. Vol. 3124 of Lecture Notes in Computer Science, edited by Jos de Souza, Petre Dini, and Pascal Lorenz, 33–44. Berlin/Heidelberg: Springer. ISBN 978-3-540-22571-3. [CrossRef]
  • 24. Hayes, J. 1968. “Adaptive Feedback Communications.” IEEE Transactions on Communications and Technology 16 (1): 29–34. ISSN 0018-9332. doi: 10.1109/TCOM.1968.1089811. [CrossRef], [Web of Science ®]
  • 25. Hole, K. J., H. Holm, and G. E. Oien. 2000. “Adaptive Multidimensional Coded Modulation Over Flat Fading Channels.” IEEE Journal on Selected Areas in Communications 18 (7): 1153–1158. doi: 10.1109/49.857915. [CrossRef], [Web of Science ®]
  • 26. Huang, J.-L., and S. Signell. 2009. “On Performance of Adaptive Modulation in MIMO Systems Using Orthogonal Space Time Block Codes.” IEEE Transactions on Vehicular Technology 58 (8): 4238–4247. ISSN 0018-9545. doi: 10.1109/TVT.2009.2022475. [CrossRef], [Web of Science ®]
  • 27. Le, L., E. Hossain, and T. Le-Ngoc. 2007. “Interaction between Radio Link Level Truncated ARQ, and TCP in Multi-rate Wireless Networks: A Cross-layer Performance Analysis.” IET Communications 1 (5): 821–830.ISSN 1751-8628. doi: 10.1049/iet-com:20060193. [CrossRef], [Web of Science ®]
  • 28. Liu, J., and S. Singh. 2001. “ATCP: TCP for Mobile ad hoc Networks.” IEEE Journal on Selected Areas in Communications 19 (7): 1300–1315. doi: 10.1109/49.932698. [CrossRef], [Web of Science ®]
  • 29. Liu, Q.-W., S.-L. Zhou, and G. B. Giannakis. 2004. “Cross-layer Combining of Adaptive Modulation and Coding with Truncated ARQ Over Wireless Links.” IEEE Transactions on Wireless Communications 3 (5): 1746–1755. ISSN 1536-1276. doi: 10.1109/TWC.2004.833474. [CrossRef], [Web of Science ®]
  • 30. Liu, Q.-W., S.-L. Zhou, and G. B. Giannakis. 2004. “TCP Performance in Wireless Access with Adaptive Modulation and Coding.” Proc. IEEE Int. Conf. Communications 7: 3989–3993. doi: 10.1109/ICC.2004.1313300. [CrossRef]
  • 31. Liu, Q.-W., S.-L. Zhou, and G. B. Giannakis. 2005. “Queuing with Adaptive Modulation and Coding Over Wireless Links: Cross-layer Analysis and Design.” IEEE Transactions on Wireless Communications 4 (3): 1142–1153. ISSN 1536-1276. doi: 10.1109/TWC.2005.847005. [CrossRef], [Web of Science ®]
  • 32. McLoughlin, I. V., and H. Sirisena. 2010. “TCP/IP Link Layer Error Mitigation for MIMO Wireless Links.” Journal of Telecommunication Systems 1–12. ISSN 1018-4864.
  • 33. Ong, L.-T., M. Shikh-Bahaei, and J. A. Chambers. 2008. “Variable Rate and Variable Power MQAM System based on Bayesian Bit Error Rate and Channel Estimation Techniques.” IEEE Transactions on Communications 56 (2): 177–182. ISSN 0090-6778. doi: 10.1109/TCOMM.2008.060142. [CrossRef], [Web of Science ®]
  • 34. Paulraj, A. J., D. A. Gore, R. U. Nabar, and H. Bolcskei. 2004. “An Overview of MIMO Communications - A Key to Gigabit Wireless.” Proc. IEEE 92 (2): 198–218. ISSN 0018-9219. doi: 10.1109/JPROC.2003.821915. [CrossRef], [Web of Science ®]
  • 35. Pottie, G. J. 1995. “System Design Choices in Personal Communications.” IEEE Personal Communications 2 (5): 50–67. doi: 10.1109/98.468362. [CrossRef]
  • 36. Prakash, P. S., and I. V. McLoughlin. 2011. Analysis of Adaptive Modulation with Antenna Selection under Channel Prediction Errors. Calicut, India: In Proc. Int. Conf. Comms. and Signal Process. [CrossRef]
  • 37. Rath, H. K., A. Karandikar, and V. Sharma. May 2008. “Adaptive Modulation-based TCP-aware Uplink Scheduling in IEEE 802.16 Networks. In IEEE Int. Conf.” Communications 2008: 3230–3236. doi: 10.1109/ICC.2008.608. [CrossRef]
  • 38. Selvam, A., K. Kathiravan, and R. Reshmi,. 2008. “A Cross-layer TCP Protocol with Adaptive Modulation for MANETs.” Int. Conf. Signal Processing, Communications and Networking, 2008; January, 428–433. doi:10.1109/ICSCN.2008.4447232.
  • 39. Sundaresan, K., V. Anantharaman, H.-Y. Hsieh, and A. R. Sivakumar. 2005. “ATP: A Reliable Transport Protocol for ad hoc Networks.” IEEE Transactions on Mobile Computing 4 (6): 588–603. doi: 10.1109/TMC.2005.81. [CrossRef], [Web of Science ®]
  • 40. Telatar, E. 1999. “Capacity of Multi-antenna Gaussian Channels.” European Transactions on Telecommunications 10 (6): 585–595. ISSN 1541-8251. doi: 10.1002/ett.4460100604. [CrossRef], [Web of Science ®]
  • 41. Webb, W. T., and R. Steele. 1995. “Variable Rate QAM for Mobile Radio.” IEEE Transactions on Communications 43 (7): 2223–2230. ISSN 0090-6778. doi: 10.1109/26.392965. [CrossRef], [Web of Science ®]

Librarians

Taylor & Francis Group