Network Coding with Multimedia Transmission and Cognitive Networking: An Implementation based on Software-Defined Radio

Nội dung text: Network Coding with Multimedia Transmission and Cognitive Networking: An Implementation based on Software-Defined Radio

1. 72 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 Invited Article Network Coding with Multimedia Transmission and Cognitive Networking: An Implementation based on Software-Defined Radio Tran Thi Thuy Quynh1, Ngo Khac Hoang1,2, Nguyen Van Ly1,3, Nguyen Linh Trung1, Nguyen Quoc Tuan1, Ejder Bastug4, Sylvain Azarian5, Le Vu Ha1, Vo Nguyen Quoc Bao6, Tran Xuan Nam7, Mộrouane Debbah8, Pierre Duhamel9 1 University of Engineering and Technology, Vietnam National University, Hanoi, Vietnam 2 Department of Electrical Engineering, Chalmers University of Technology, 41296 Gothenburg, Sweden 3 Computational Science Research Center, San Diego State University, CA, USA 4 Nokia Bell Labs, 91620 Nozay, France 5 SDR-Technologies Ltd., France 6 Posts and Telecommunications Institute of Technology, Ho Chi Minh City, Vietnam 7 Faculty of Radio Electronics, Le Quy Don Technical University, Hanoi, Vietnam 8 Huawei Lagrange Mathematical & Computing Research Center, 75007 Paris, France 9 Signals and Systems Laboratory, CentraleSupelec, Paris-Saclay University, 91190 Gif-sur-Yvette, France Correspondence: Nguyen Linh Trung, linhtrung@vnu.edu.vn Communication: received 29 December 2020, revised 31 December 2020, accepted 31 December 2020 Online publication: 25 February 2021, Digital Object Identifier: 10.21553/rev-jec.263 The associate editor coordinating the review of this article and recommending it for publication was Prof. Nguyen Tan Hung. Abstract– Network coding (NC) is considered a breakthrough to improve throughput, robustness, and security of wireless networks. Although the theoretical aspects of NC have been extensively investigated, there have been only few experiments with pure NC schematics. This paper presents an implementation of NC under a two-way relay model and extends it to two non-straightforward scenarios: (i) multimedia transmission with layered coding and multiple-description coding, and (ii) cognitive radio with Vandermonde frequency division multiplexing (VFDM). The implementation is in real time and based on software-defined radio (SDR). The experimental results show that, by combining NC and source coding, we can control the quality of the received multimedia content in an on-demand manner. Whereas in the VFDM-based cognitive radio, the quality of the received content in the primary receiver is low (due to imperfect channel estimation) yet retrievable. Our implementation results serve as a proof for the practicability of network coding in relevant applications. Keywords– Network coding, two-way relay model, software-defined radio (SDR), orthogonal frequency division multiplexing (OFDM), Vandermonde frequency division multiplexing (VFDM), multimedia, cognitive radio. 1 Introduction network topology. In this model, two end nodes A and B expect to exchange their respective data packet a and In 2000, network coding (NC) was first introduced b with each other via a relay node R. It is assumed that by Ahlswede in [1] to improve network throughput. there is no direct link between A and B due to, e.g., Unlike the store-and-forward mechanism in traditional blockage and limited radio range. We consider three routing protocols, NC allows intermediate nodes be- communication schemes. tween source and destination to perform additional First, following a traditional store-and-forward computations (coding) on the incoming data before scheme, the network needs four time slots to com- forwarding the coded information. plete the packet exchange, as illustrated in Figure 1(b). There are two main NC schemes, namely straightfor- Specifically, A transmits packet a to R in the first time ward network coding (SNC) where coding is performed slot, and R forwards a to B in the second time slot. on digital bit streams after they have been received [2], Then, B transmits packet b to R in the third time slot, and physical-layer network coding (PNC) where coding and R forwards b to A in the fourth time slot. Note that is performed via the additive nature of simultaneously node R merely forward the packets without processing arriving electromagnetic waves [3, 4]. The latter was the data contents. shown to achieve a higher throughput performance as Second, following SNC, as illustrated in Figure 1(c), illustrated in the following example. the number of time slots can be reduced. The relay node Let us consider a two-way relay (TWR) model de- R first receives sequentially both packets a and b, then picted in Figure 1(a), which is a simple and popular performs the bit-wise exclusive OR (XOR) operation, 1859-378X–2020-3402 â 2020 REV
2. T. T. T. Quynh et al.: Network Coding with Multimedia Transmission and Cognitive Networking 73 ceived signals are preferred since they yield significant performance improvements. The first successful implementation of PNC was re- ported in [7]. However, this system operates in an offline manner. The first real-time PNC implementation was introduced in [8], based on USRP N210 with XCVR2450 boards. In this system, frequency division duplex (FDD) was employed to separate the uplink and the downlink. To overcome some implementation challenges such as synchronization, packet detection, (a) TWR model (b) Conventional: 4 time slots channel estimation, and carrier-frequency-offset (CFO) estimation, the authors used orthogonal frequency di- vision multiplexing (OFDM) with beacons. In addition, PHY-layer forward error control and MAC-layer ARQ error control were also used to improve reliability. However, this implementation requires a change in the frame format and a balance between the power of the end nodes’ signals received at the relay. Another implementation of NC was presented in [9]. (c) SNC: 3 time slots (d) PNC: 2 time slots This prototype is for SNC and half-duplex packet switching, also based on USRP with RFX2400 daughter- Figure 1. Conventional forwarding and network coding methods in boards. Some other recent implementations of network the TWR model. coding can be found in, for example, [10, 11]. In this paper, we present an SDR-based implementa- tion of NC under the TWR model and two extended denoted by , over these packets to produce a new ⊕ network models for multimedia transmission and cog- single packet a b. In the third time slot, node R sim- ⊕ nitive radio, respectively. Our main contributions are ply broadcasts this coded packet. The two end nodes three-fold. First, we implement NC under the TWR can recover their expected packet based on their own model operating in full-duplex transmission mode. In packet and the received coded packet. Specifically, A our scheme, we let node A and node B transmit in can recover b as b = a (a b) and B can recover a as ⊕ ⊕ separated frequencies in the first time slot, and the relay a = b (a b). In this way, SNC reduces the number ⊕ ⊕ node performs the XOR operation on digital bit streams of time slots to three, achieving a 33% throughput gain. and then broadcasts the XOR-ed packet in the second Finally, following PNC, illustrated in Figure 1(d), time slot. Compared to existing implementations of A and B transmit simultaneously. Their packets are NC under the TWR model, our scheme combines the combined on the air due to the additive nature of the advantages of both SNC and PNC; that is, it requires electromagnetic waves. Node R properly processes the only two time slots while avoiding complex PNC map- received signal in order to produce the coded packet ping. Furthermore, the proposed scheme can be easily before forwarding in the second time slot. Therefore, integrated into existing systems since it does not require PNC requires only two time slots, achieving a 50% any change in the frame format. We also implement an throughput gain compared to the traditional store-and- error correction code to improve the reliability. Second, forward routing scheme. we present the first implementation of NC for multime- Although NC has been widely analyzed and as- dia transmission with joint source-network coding. Two sessed via both theoretical analyses and numerical sim- types of source coding are considered: layered coding ulations, limited results have been obtained via real- and multiple-description coding. NC is employed to channel implementation. One of the first implementa- ensure that the destination can still receive data packets tions of NC was reported in [5], where a simplified when the direct link is lost. Finally, we present the version of PNC, called analog network coding (ANC), first implementation of NC for a special cognitive radio was introduced. network based on Vandermonde frequency division The idea of ANC is that the relay node simply multiplexing (VFDM). Specifically, we consider a TWR amplifies then retransmits the received superimposed model in the secondary tier of an overlay two-tier signals without coding. This implementation was based network. To avoid cross-tier interference to the primary on a software-defined radio (SDR) platform consisting receiver, the relay node uses VFDM to transmit signals of the Universal Software Radio Peripheral (USRP) with on the nulls of the primary channel. RFX2400 daughterboards as the hardware and the open The rest of the paper is organized as follows. Sec- source GNU Radio [6] as the signal processing software. tion 2 describes our NC framework and prototype, Although ANC is simple to implement, it has a as well as three scenarios whose implementations are critical drawback of error propagation since the relay subsequently presented. Specifically, Section 3 presents amplifies the noise along with the signal before for- the implementation of NC under the TWR model. warding. PNC systems in which the relay performs Section 4 extends the implementation of NC, combining XOR or other denoising PNC mappings on the re- source coding with multimedia transmission. Section 5
3. Two-way Relay Model NODE A OFDM osmocom Message M1 Transmitter Sink NODE B OFDM osmocom Message M1 Transmitter Sink NODE R Frequency Lowpass OFDM Message M1 Mixer Filter Receiver osmocom OFDM osmocom Source XOR +β Transmitter Sink Frequency Lowpass OFDM Message M2 Mixer Filter Receiver -β Cognitive Network Coding NODE SR NODE A/B osmocom VFDM OFDM osmocom Source Receiver Transmitter Sink osmocom OFDM OFDM Hybrid Stream osmocom Secondary Source 1 Receiver data Transmitter Endnode Controller Sink Controller channel Primary Source Secondary Secondary Controller Data Sink Data Source Primary Data Source channel VFDM Metrics data VFDM Controller channel Transmitter NODE D Frequency OFDM Shifter Receiver data osmocom Secondary 74 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 Source 2 +β Relay Controller presents the implementation of NC in the VFDM-based osmocom OFDM OFDM osmocom cognitive radio network. Finally, Section 6 gives con- Source Receiver Transmitter Sink Frequency OFDM Shifter Receiver cluding remarks. Primary Destination 2 Proposed Network Coding Framework Controller and Prototype Primary Data -β Sink 2.1 Proposed Network Coding Framework Following the principle of NC, we present a frame- work in which relay nodes are inserted in the network to detect and exploit coding opportunities to forward (a) A 4-node network model multiple packets in a single transmission in order to Decode as OFDM Transmit OFDM Transmit OFDM Decode as OFDM improve network throughput and reliability. Consider S D S D a general mesh network of which each node has packets hSD S D S D f1 to transmit and would like to receive packets from sS yD f2 certain other nodes. Our considered framework incor- porates the following features. f1 f2 Relaying infrastructure: Relay nodes are deployed in Transmit OFDM Transmit OFDM Decode as VFDM Decode as VFDM hRD the network to help forward the packets. These nodes f1 f1 are placed at positions where there is no direct link sA sB f0-β f0+β A R B A R B sR between the communicating nodes due to blockage Decode as OFDM and Transmit VFDM or limited radio range, or specially designed so as to A R B A R B Estimate channel maximize the coding gain. Opportunistic listening: Exploiting the broadcast na- (b) A 2-tier cognitive radio network ture of wireless media, the relay nodes opportunisti- Figure 2. Two scenarios of interest for NC implementation, apart from cally listen to the packets transmitted from surrounding the TWRx model.S pilots nodes. It is assumed that each node includes its identity OFDM TX OFDM TX and the identities of the other nodes from which it pilots would like to receive packets. Thus, the relay nodes relay nodesACK to simultaneously receive theSDR packets, Device the SDR Device ACK can detect this information from the packets. communicating nodes employ FDD. Specifically, they Opportunistic coding: From knowledge of the desired shift their carrier frequencyOFDM RX by a small amount from OFDM RX yS packets of each node, each relay node identifies the a central frequency using a frequency mixer, so that opportunity to encode a subset of the received pack- their signals are separated. The relay nodes receive at ets to form a coded packet and broadcast this coded the central frequency andS sample with a sufficiently D packet. Since there can be multiple coding options cor- wide bandwidth so as to recover the packets of multiple responding to multiple subsets of the received packets, nodes. Third, we use an SDR platform that consists the coding operation should be well designed. On one of bladeRF hardware [12] and GNU Radio software. pilots hand, the relay nodes should maximize the number of Each node is a commodity personal computer running OFDM TX xA single packets encoded to reduce the transmission time. GNU Radio and connected to one or more bladeRF- On the other hand, these coded packets should enable pilots SDR Device ACK x115 devices. The bladeRF devices operate in the full- XOR the intended nodes to decode their desired packets. duplex mode withV twoFD associatedM TX VERT 2450 antennas yB For example, under the TWR model introduced in for transmission and reception. VFDM RX Section 1, after the first phase, the relay has received xR We consider three scenarios of NC forSDR implementa- Device yR packets a and b from nodes A and B, respectively. It tion as follows.yA The first scenario is the TWR model A further knows that node A has packet a while desiring presented in SectionOF 1,DM where RX nodes A and B would pilots packet b, and that node B has packet b while desiring like to exchange packets with the help of a relay. The OFDM TX packet a. Therefore, a coding opportunity occurs, and second scenario,XOR shown in Figure 2(a), is a 4-node xB the optimal coding operation is to combine a and b network where nodes A and B would like to send SDR Device ACK and send this combination, a b. In general, a sub- packets to a destinationOFDM node, RX C, with the help of a XOR ⊕ yB yA set of received packets should be combined if all the relay. We assume that the packets from nodes A and B VFDM RX intended nodes already have all packets in the subset are generated from a specific type of source coding for but one packet that they desire. Note that the number of multimedia transmission.R The third scenario, shown in yR encoded single packets, so-called the coding field size, Figure 2(b), is a two-tier cognitive radio network with B can be larger than 2. The field size goes up linearly with a transceiver pair at the primary tier and a TWR model the number of nodes in the network. at the secondary tier. In these scenarios, we restrict the NC operation to a field size of 2, i.e., the relay node 2.2 Proposed Prototype combine the incoming packets from two nodes. Within the described framework, we design a proto- We would like to emphasize that our prototype is the type with the following elements. First, the nodes trans- first case study of the interplay among network coding, mit their packets using OFDM. Second, in order for the source coding, OFDM and cognitive radio.
4. T. T. T. Quynh et al.: Network Coding with Multimedia Transmission and Cognitive Networking 75 Figure 3. A system block diagram of the TWR model. In our implementation, we use the version 3.7.6 of GNU Radio running in a dual-core general purpose processor under Ubuntu 14.04 OS. Experiments take place in a 25 m2-area closed laboratory. We place the devices at least 0.5 m away from each other. We make use of the OFDM tranceiver blocks developed in the gr- s4a module [13]. Further details about the experimental setup are given in each following section, where we Figure 4. Frequency allocation in the TWR model. present sequentially the implementation of the three aforementioned scenarios. 3.1.1 Separation of Uplink Signals: Figure 4 illustrates the frequency allocation in our im- 3 Implementation of Network Coding plementation. Let f0 be a (high) carrier frequency. Node under the TWR Model A transmits on frequency f0 β while node B on fre- − quency f0 + β. Node R transmits on another frequency In this section, we focus on an SDR-based real-time f1. Node R receives on frequency f0, but samples with implementation of NC under the TWR model. a wide bandwidth enough to completely receive both f β and f + β so that node R can retrieve signals 0 − 0 transmitted from nodes A and B separately. 3.1 System Model Figure 3 depicts a signal processing block diagram of We consider the TWR model described in Section 1. the TWR model. In particular, at node R, the input sig- To obtain reliable transmission, we need to overcome nal received at f0 goes through two separate branches. several challenges, in particular: (i) separation of up- In one branch, the input signal is shifted by an amount link signals at the relay node, (ii) time and frequency of β Hz at a frequency mixer, and then the mixed signal synchronization, and (iii) channel estimation. is low-pass filtered to retrieve the signal transmitted by Our solutions respectively are: (i) use frequency node A. Similarly, in the other branch, the input signal division multiplexing to distinguish uplink signals is shifted by β Hz at another frequency mixer, and at the relay node, (ii) employ OFDM and exploit then the mixed− signal is low-pass filtered to retrieve the the preamble part for time and frequency synchro- signal transmitted by node B. The messages obtained nization following the Schmidl-Cox synchronization in the two branches after OFDM demodulation are then method, and (iii) insert pilot symbols in an OFDM combined into a new message by the XOR operation. frame for channel estimation. Next, we will describe This new message is then modulated and broadcast these solutions. to A and B.
5. 76 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 (a) Transmited image at node A (b) Transmited image at node B Figure 5. System operating mechanism. 3.1.2 Communication Protocol: Our system works in sessions. The transmission pro- tocol within each session is illustrated in Figure 5. The relay node first broadcasts a beacon message to tell the two end nodes the start of a session. When a session starts, each end node loads N native packets and stores them in a buffer. After that, a checking index i runs (c) Received image at node A, (d) Received image at node B, from 0 to N 1. At each value of i, the end node checks BER = 0.0128 BER = 0.0122 − whether it has received the corresponding i-th XOR-ed Figure 6. Transmitted and received images under the implemented packet from the relay or not. If yes, the checking index TWR model. increases one. If no, the end node transmits the i-th native packet and then i is increased by one. If i = N, but the end node has not received all N XOR-ed packets 3.2 SDR Implementation and Results yet, it will be returned to zero (i = 0). Of course, for As mentioned earlier, we make use of the gr-s4a the first run of the index i through the buffer, the end module [13] for OFDM modulation and demodulation. node certainly has to send all the loaded native packets. To make the system work with the operating mech- Thus, this protocol allows the end nodes to proceed anism described above, we develop some controller to the transmission of the next native packet without blocks for the two end nodes and the relay node. We having to wait for the successful transmission of the demonstrate the exchange of two images between the corresponding XOR-ed packet from the relay. two end nodes. Each end node transmits a 256 256 At the relay node, whenever it receives a native image. An instance of transmitted and received imagesì packet from an end node, it will check whether the is shown in Figure 6. It is observed that the transmitted corresponding native packet from the other end node images from a node were well retrieved at the other 2 is received. If yes, and the XOR-ed packet has not node with a bit error rate (BER) of around 10− . been created yet, the relay node will combine the two corresponding native packets into a XOR-ed packet and store it in a buffer. If no, the received native packet is 4 Implementation of Joint just stored in a buffer. The XOR-ed packet is broadcast Source-Network Coding when it is available. A new session starts whenever both end nodes have received all N XOR-ed packets. Based on the implementation of NC under the TWR 3.1.3 Error Correction Code: model described in Section 3, we extend the imple- To improve the communication reliability, we de- mentation to a 4-node network with joint source- ploy a Hamming (7,4) code with the following network coding to show the usefulness of NC for generator matrix: multimedia transmission. We consider two types of source coding. The first   1 1 1 0 0 0 0 type is layered coding (LC), which is widely used in 1 0 0 1 1 0 0 multimedia source coding. It generates one base layer   G =  . (1) and some n enhanced layers. The base layer is the 0 1 0 1 0 1 0 most important layer and essential for the data stream 1 1 0 1 0 0 1 to be recovered. Without receiving the base layer, the This means that three redundant bits are inserted in data stream cannot be recovered since the use of other the first, second and fourth positions. A 4-bit message enhanced layers depends on the content of the base is encoded into a 7-bit codeword. At the receiver, the layer. The enhanced layers are to improve the quality decoder calculates the syndrome metric and estimates of the data stream. However, the first enhanced layer the transmitted message. This Hamming code is able depends on the base layer and each enhanced layer to correct one bit error. n + 1 depends on enhanced layer n. Thus a certain
6. T. T. T. Quynh et al.: Network Coding with Multimedia Transmission and Cognitive Networking 77 (a) Direct link A-C is lost (a) Direct link A-C is lost (b) Direct link B-C is lost (b) Direct link B-C is lost Figure 7. 4-node network model with conventional relay. Figure 8. 4-node network model with NC. layer n can only be applied if n 1 layers were already − illustrated in Figure 7. The links A-R, B-R, and R-C are applied. Hence, data streams using this layer coding supposed to be stable. It can be seen that, thanks to approach can be interrupted whenever the base layer the addition of the relaying station (node R), C can still is missing. receive packets transmitted from A and B even when The second type of source coding is multiple- one of the two direct-links is lost because when node description coding (MDC) in which the data stream is R is active, it relays every packet it receives to C. divided into n independent sub-streams (n 2), these ≥ Now, consider the 4-node network model with NC sub-streams are called descriptions. Thus, MDC is a as shown in Figure 8. Node R will perform NC on two form of data partitioning. The packets of a description packets it received (a and b) to create a new packet, can be sent over different paths. In contrast to LC where which is a b, and then forward this new packet to C. the use of layer n depends on layer n 1, every received ⊕ − Suppose that the link between A and C (A-C) is lost as description of MDC at the destination can be used to in Figure 8(a). At node C, based on the packet b received recover the original data stream. This means that the directly from B and the XOR-ed packet received from quality of the decoded stream is proportional to the R, the packet a can be recovered as a = b (a b). ⊕ ⊕ number of received descriptions. Since any received Similarly, if the link B-C is lost, b can be recovered as description can be used for the decoding process, the b = a (a b). Thus, if one of the two direct-links ⊕ ⊕ data stream is rarely interrupted (except for link loss). is lost, using NC, node R simply relays the XOR-ed The loss of some descriptions only results in reduced packet to C without knowing which link is lost and still quality of the decoded stream. ensures that C can recover both a and b. Recall that for the case of using the conventional relaying mechanism, 4.1 System Model node R has to transmit both a and b since it does not know which direct-link is lost. 4.1.1 A 4-node Network Coding System Model: 4.1.2 A 4-node Joint Source-Network Coding Model: We consider a wireless network model with 4 nodes Now, we combine source coding (at A and B) with NC illustrated in Figure 2(a), in which nodes A and B are (at R), as shown in Figure 9. two source nodes, node C is the destination, and node We assume that the direct-link B-C is lost. Each R is the relay. Both A and B want to send data to C source node (A or B) transmits a layer (or a description). and both have the direct link to C. Node R works as Node performs NC over the two received packets ( a relaying station with the aim of assisting the data R a and b) to create a new coded packet c as follows: transmission of A and B. Node R will relay all of its received packets to node C. The presence of node R c = a βb, (2) in the system is to improve the possibility of receiving ⊕ where β 0, 1 . We consider the following two cases: data packets at C in case a direct link is lost between ∈ { } A and C (link A-C) or between B and C (link B-C). Case 1: Node R does not have any information • Consider the situation in which the above 4-node net- about packet b, meaning that b is considered as work model employs only traditional relaying. Suppose a normal data packet, β is set to be 0 or 1 with that one of the two direct-links (A-C or B-C) is lost, as equal probabilities.
7. 78 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 Figure 9. NC with source coding in 4-node network model. (a) Decoded image using 1 de- (b) Decoded image using 2 de- scription, BER = 0.1269 scriptions, BER = 0.000203 Figure 10. Frequency allocation in 4-node network model. (c) Decoded image without in- (d) Decoded image with in- formation about source coding, formation about source coding, BER = 0.2673 BER = 0.0108 Case 2: Node R has information about packet b, • Figure 11. Decoded images by LC and MDC. i.e., R knows if the packet b is of a layer or a description. Essential for the decoding process at C, the parameter β is set to be 1. This is to give by filtering the image with a lowpass filter, and the priority to packets transmitted from B. enhanced layer is generated by having the original Figure 10 illustrates the frequency allocation of this image subtracted by the base layer. We build a block 4-node network model. Here we still suppose that the in GNU Radio for decoding at the destination so that direct-link B-C is lost. The two source nodes A and B the image can be recovered directly in GNU Radio. transmit on frequencies f1 and f2, respectively. Node Similarly, for MDC, we implement MDC with only two R receives on f1 and f2, and transmits on f3. Since the descriptions. Received descriptions at the destination link B-C is supposed to be lost, node C can only receive are used to recover the original image. This is done by signals on f1 and f3. In addition, node C makes use of a block in GNU Radio. For the case of LC, A transmits a controlling channel f4 to transmit control messages the enhanced layer and node B transmits the base layer. to A and B. Packets transmitted from A and B will For the case of MDC, node A transmits one description be combined into a XOR-ed packet to be relayed on f3. and node B transmits the other description. Each source All nodes in the network apply OFDM modulation and node is a commodity PC connected to a bladeRF device. demodulation techniques. For two nodes R and C, each node is a PC connected This 4-node network also works in sessions. A session to two bladeRF devices. starts when node C sends a control message on f4 to Experimental results are shown in Figure 11. It is nodes A and B. Therefore, A and B know when to start clearly seen that the decoded image quality is increased a session and send their data. Whenever received the as more descriptions/information about source coding control message, end nodes (A, B) will load N packets are used in decoding. and then store them in a buffer. After that, end nodes will send N packets continuously until receiving the next control message for the next session. 5 Implementation of VFDM-Based Cognitive Radio with Network Coding 4.2 SDR Implementation and Results Cognitive radio has been attracting a sustained at- In our implementation, LC/MDC is first performed tention for its potential to improve spectral efficiency. in Matlab to generate text files containing the lay- Cognitive radio enables the deployment of a two-tier ers/descriptions. Then, the controller block of source network, composed of a primary tier and a secondary nodes in GNU radio software loads a text file cor- tier. The former is licensed to use a specific spectrum responding to a layer/description and sends it. For range, whereas the latter accesses this spectrum to carry simplicity, we implement LC with only two layers (the their transmission without interfering primary trans- base layer and one enhanced layer). The data to be missions. To avoid cross-tier interference to primary coded is a grayscale image. The base layer is generated users, secondary users must adapt their transmission
8. Two-way Relay Model NODE A OFDM osmocom Message M1 Transmitter Sink NODE B OFDM osmocom Message M1 Transmitter Sink NODE R Frequency Lowpass OFDM Message M1 Mixer Filter Receiver osmocom OFDM osmocom Source XOR +β Transmitter Sink Frequency Lowpass OFDM Message M2 Mixer Filter Receiver -β Cognitive Network Coding NODE SR NODE A/B osmocom VFDM OFDM osmocom Source Receiver Transmitter Sink osmocom OFDM OFDM Hybrid Stream osmocom Secondary Source 1 Receiver data Transmitter Endnode Controller Sink Controller channel Primary Source Secondary Secondary Controller Data Sink Data Source Primary Data Source channel VFDM Metrics data VFDM Controller channel Transmitter NODE D Frequency OFDM Shifter Receiver data osmocom Secondary Source 2 +β Relay Controller osmocom OFDM OFDM osmocom Source Receiver Transmitter Sink Frequency OFDM Shifter Receiver Primary Destination Controller Primary Data -β Sink Decode as OFDM Transmit OFDM Transmit OFDM Decode as OFDM S D S D hSD hSD S D S D f1 yD sS yD sS f2 hRD f1 f2 Transmit OFDM Transmit OFDM Decode as VFDM Decode as VFDM hRD sR f1 f1 B sA sB sA f0-β f0+β s A R B A R B sR Decode as OFDM and Transmit VFDM A R B A R B Estimate channel T. T. T. Quynh et al.: Network Coding with Multimedia Transmission and Cognitive Networking 79 S D sS pilots hSD OFDM TX OFDM TX sS yD pilots SDR Device SDR Device ACK hSB ACK hSA hRD OFDM RX OFDM RX s’S yA hRA hRB yB S D sR A R B pilots Figure 12. A transmission phase of the cognitive radio network. Figure 13. A 4-node hybrid cognitive radio network model. OFDM TX sA pilots SDR Device ACK under some constraints in terms of signal power and crete Fourier transform (DFT) matrix with [F]k+1,l+1 = XOR channel access time. However, these constraints make 1 i2π kl VFDM TX s''B e− N , k, l = [0, 1, , N 1], A the CP insertion VFDM RX reliable transmission challenging for secondary users. √N − matrix of size (N + L) N, E C(N+L) L the VFDM sR To guarantee an acceptable reliability for the secondary ì SDR Device s'R ì ∈ A transmissions, NC, with its potential to improve the precoder matrix. s'A The precoded transmit vector x and x CN+L at OFDM RX pilots network throughput and shorten the transmission time, S R ∈ S and R, respectively, are defined as follows: OFDM TX can be applied at the secondary tier. XOR sB Based on the overlay approach, VFDM was proposed 1 xS = AF− sS (3) in [14] as a technique for cross-tier interference manage- SDR Device ACK x = Es . (4) OFDM RX XOR ment. With VFDM, a secondary transmitter uses linear R R s'B s''A   VFDM RX precoding to project its signal onto the null space of Let hMN = hMN,0, , hMN,L be the (L + 1)-tap the interfering channel from the secondary transmitter fading channel vector, modeling the downlink between R s'R B to the primary receiver. Therefore, the secondary trans- the transmitter M and the receiver N. The convolution mitter can transmit over the same band as the primary of each precoded symbol vector xM with corresponding transmitter but does not cause any interference at the channel hMN can be modeled as a Toeplitz matrix (N+L) (N+L) deskPDF Studio Trial primary receiver. HMN C ì . The received signal yD N ∈ N+L N+L ∈ There have been some results on VFDM implemen- C , yA C and yB C at D, A and B, tation in a VFDM standalone transceiver pair [15] or are respectively∈ ∈ a two transmitter – two receiver scenario, where both
   transmitters are implemented on the same baseband yD = FB HSDxS + HRDxR + nD (5) transceiver, called hybrid transceiver [16]. This hybrid yA = HSAxS + HRAxR + nA (6) transceiver approach for cognitive network deployment yB = HSBxS + HRBxR + nB (7) was proposed in [17]. where B = [0N LIN] is the CP removal matrix, nM is In this section, we present an implementation of ì a two-tier cognitive network adopting NC at the the Gaussian noise vector at receiver M A, B, D 2 ∈ { } secondary tier. with covariance matrix σ I. By analyzing the interference constraint that R must satisfy, E can be built so that no interference signal component is perceived at D after the CP removal 5.1 System Model and DFT. Let H˜ RD = FBHRD. For a zero cross-tier Consider a 5-node two-tier network scenario, as interference, the following condition must be satisfied: shown in Figure 2(b), where a primary system com- ˜ ˜ posed of a transceiver pair denoted by S (source)/D HRDxR = HRDEsR = 0N. (8) (destination), shares the spectrum with an opportunis- This is achieved by a special Vandermonde matrix tic secondary system which is a TWR network com- construction of the linear procoder E, built from the posed of two end nodes A, B and one relay node R. roots of the polynomial S(z) with L + 1 coefficients of The primary system communicates a message s over L 1 S the interference channel HRD as S(z) = ∑i=0 hRD,iz− . the licensed frequency band, whereas the secondary system access this spectrum to exchange two messages 5.2 SDR Implementation sA (from A) and sB (from B) between two end nodes via the relay node. 5.2.1 Baseband design: Consider the first phase of this network depicted in We consider the hybrid transceiver approach [17]. The Figure 12, where S performs an OFDM transmission primary transmitter S and the secondary relay R are im- towards D with N subcarriers and a cyclic prefix (CP) of plemented on one baseband transceiver. Consequently, size L. The total block length is N + L. Node R performs the aforementioned 5-node network model is trans- a VFDM transmission with the same size of N + L. formed into a 4-node hybrid network model as shown Let s CN, s CL be the transmit symbol vector in Figure 13. The hybrid transceiver is denoted as SR. In S ∈ R ∈ at S and R, respectively, F CN N the unitary dis- the implementation test-bed, it is a PC connected to two ∈ ì
9. 80 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 80 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 (a) Node SR (a) Node SR (b) Node D (b) Node D (c) Nodes A, B (c) Nodes A, B Figure 14. Signal processing chains of nodes in the considered cognitive radio network. Figure 14. Signal processing chains of nodes in the considered cognitive radio network. Table I bladeRF devices. The main motivation for this hybrid Table I Operating Mode and Frequency Allocated of Nodes for link can be directly used to build the precoding matrix approach is that the channel estimate of the primary Operating ModeUplink and F andrequencyDownlinkAllocatedPhases of Nodes for Uplink and Downlink Phases E for the VFDM transmission. link can be directly used to build the precoding matrix Each node operates following a specific configura- E for the VFDM transmission. Node Uplink Downlink tion, depending on its transmitting/receiving mode in SR OFDM RX, f OFDM TX + VFDM TX, f uplink/downlink phase and its role in the network. The Each node operates following a specific configura- 1 1 operating modes and frequency allocated of each node D OFDM TX, f1 OFDM RX, f1 tion, depending on its transmitting/receiving mode in are given in Table I. A OFDM TX, f0 β VFDM RX, f uplink/downlink phase and its role in the network. The − 1 An illustration of the signal processing chain in each operating modes and frequency allocated of each node B OFDM TX, f1 + β VFDM RX, f1 node is shown in Figure 14. The main algorithms are given in Table I. for cognitive transmission strategy are embedded in controller blocks (in green). An illustration of the signal processing chain in each bladeRF5.2.2 Communication devices. The main Protocol: motivation for this hybrid 5.2.2 Communication Protocol: node is shown in Figure 14. The main algorithms Weapproach hereafter is that describe the channel simply estimate the adopted of the primary protocol inWe hereafter describe simply the adopted protocol in for cognitive transmission strategy are embedded in the order of channel access. The network operates controller blocks (in green). under the channel reciprocity principle in TDD mode.
10. T. T. T. Quynh et al.: Network Coding with Multimedia Transmission and Cognitive Networking 81 In the uplink phase, D sends a message sS to SR at the center frequency f1. This can be either a beacon message to trigger the communication or an ACK mes- sage to confirm the successful reception from SR in the previous downlink phase. In both cases, pilot symbols are added to enable the estimation of HDR at SR. At SR, thanks to the principle of channel reciprocity, HDR can be used as HRD by the VFDM transmitter to construct the precoder E. Meanwhile, in the secondary tier, the end nodes A and B send their message s and A (a) Transmitted image from SR (b) Decoded image at D sB, respectively, to SR at the center frequency f0 + β and f0 β, respectively. These signals are carried by Figure 15. The transmitted (128 128)-pixel and decoded image in ì frequencies− different from that carrying the signal from the primary tier of the implemented cognitive radio network. D and thus cause no interference to the latter. Node SR receives the primary signal from D at center fre- deskPDF Studio Trial quency f1 and receives the secondary signals at center NC in the secondary tier. Our implementation test- frequency f0 with wide-enough sampling bandwidth to beds operate in real time using the SDR technology. capture both signals from A and B. Our results not only complement the theoretical anal- In the downlink phase, as soon as SR receives the yses in the literature in showing that NC can im- signal from D, the downlink phase starts. Both OFDM prove the network throughput and shorten transmis- frame containing the primary data for primary trans- sion time, but also confirm the compatibility of NC in mission and VFDM frame containing XOR-ed version practical applications. of received messages from A and B for secondary trans- mission are generated at SR. Two frames are combined for a hybrid frame to be transmitted at f1. All the Acknowledgment receiving nodes D, A and B receive this frame but demodulate in their custom mode (OFDM or VFDM) This work was partly supported by the Ministry of to extract their intended data. To be more specific, Science and Technology of Vietnam, under project num- D demodulates in the OFDM mode while A and B ber 39/2012/HD/NDT, and by the French DIGITEO demodulate in the VFDM mode. Afterwards, an ACK organization. The authors would like to thank Prof. message containing a pilot is generated at D to be Muriel Mộdard from the Department of Electrical Engi- transmitted back to SR in the next uplink phase in order neering and Computer Science, Massachusetts Institute to acknowledge the successful reception and trigger the of Technology, for her invaluable comments to help us next session. The secondary end nodes retrieve the data improve the perspective of the paper. from the other end node from the received XOR-ed message from SR and its original data. References 5.3 Experimental Results [1] R. Ahlswede, N. Cai, S.-Y. R. Li, and R. W. Yeung, “Net- In the experiment, we consider carrier frequencies work information flow,” IEEE Transactions on Information f = 2.435 GHz, f = 2.415 GHz, and frequency Theory, vol. 46, no. 4, pp. 1204–1216, 2000. 1 0 [2] C. Fragouli, Y. Boudec, J, and J. Widmer, “Network cod- offset β = 250 KHz. The sampling bandwidth is 1 ing: An instant primer,” in SIGCOMM Comput. Commun. MHz. We let node SR send a grayscale (128 128)- Rev., vol. 36, no. 1. ACM, Jan. 2006, pp. 63–68. pixel image and focus on the primary transmissionì [3] S. Zhang, S. C. Liew, and P. P. Lam, “Hot topic: Physical- result. In Figure 15, we show the transmitted image layer network coding,” in Proceedings of the 12th Annual from SR and decoded image at D. The received image international conference on Mobile computing and network- ing. ACM, 2006, pp. 358–365. at D is at low quality (due to the imperfect channel [4] S. C. Liew, S. Zhang, and L. Lu, “Physical-layer network estimation at SR) but observable. This first observation coding: Tutorial, survey, and beyond,” Physical Commu- shows that the primary system can still communicate its nication, vol. 6, pp. 4–42, 2013. information while allowing the VFDM-based secondary [5] S. Katti, S. Gollakota, and D. Katabi, “Embracing wireless system access its spectrum. Although the number of interference: Analog network coding,” in ACM SIG- COMM Computer Communication Review, vol. 37, no. 4. faulty pixels in Figure 15b is still high, our result serves ACM, 2007, pp. 397–408. as a proof for practical feasibility of NC and VFDM in [6] [Online] cognitive radio network. [7] L. Lu, T. Wang, S. C. Liew, and S. Zhang, “Imple- mentation of physical-layer network coding,” Physical Communication, vol. 6, pp. 74–87, 2013. 6 Conclusions [8] L. Lu, L. You, Q. Yang, T. Wang, M. Zhang, S. Zhang, and S. C. Liew, “Real-time implementation of physical- In this paper, we have presented an implementation of layer network coding,” in Proceedings of the 2nd Workshop on Software Radio Implementation Forum. ACM, 2013, pp. OFDM-based NC in three scenarios, namely, a TWR 71–76. model, a joint source-network coding for multimedia [9] D. Kramarev, Y. Hong, and E. Viterbo, “Software defined transmission, and a cognitive radio network adopting radio implementation of a two-way relay network with
11. 82 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 digital network coding,” in Australian Communications Nguyen Van Ly received the B.Eng. degree in Theory Workshop (AusCTW). IEEE, 2014, pp. 120–125. electronics and telecommunications from Uni- [10] M. Taghouti, M. Tomoskozi, M. Howeler, D. E. Lucani, versity of Engineering and Technology (VNU- F. H. P. Fitzek, A. Bouallegue, and P. Ekler, “Implemen- UET), Vietnam National University, Hanoi, tation of network coding with recoding for unequal- Vietnam, in 2014, and the M.Sc. degree in advanced wireless communications systems sized and header compressed traffic,” in IEEE Wireless from CentraleSupộlec, Paris-Saclay University, Communications and Networking Conference (WCNC), 2019, France, in 2016. He is currently pursuing pp. 1–7. the Ph.D. degree in a joint doctoral pro- [11] S. Kafaie, Y. P. Chen, O. A. Dobre, and M. H. Ahmed, gram in computational science with San Diego “Network coding implementation details: A guidance State University and University of California, document,” arXiv preprint arXiv:1801.02120, 2018. Irvine, CA, USA, and also an adjunct researcher at the Advanced [12] [Online] Institute of Engineering and Technology of VNU-UET. He received a [13] E. Bastug, “Study of Vandermonde frequency division Best Paper Award from the IEEE International Conference on Com- multiplexing on software defined radio platform,” Mas- munications (ICC) in 2020. His research interests include wireless communications, signal processing, and machine learning. ter of Science, Fatih University, Istanbul, Turkey, May 2012. [14] L. S. Cardoso, M. Kobayashi, O. Ryan, and M. Debbah, “Vandermonde frequency division multiplexing for cog- nitive radio,” in IEEE 9th Workshop on Signal Processing Advances in Wireless Communications (SPAWC), 2008, pp. 421–425. Nguyen Linh Trung obtained his B.Eng. and [15] M. Maso, L. S. Cardoso, E. Bastug, N. Linh-Trung, Ph.D. degrees, both in Electrical Engineering, M. Debbah, and O. Ozdemir, “On the practical imple- from Queensland University of Technology, Brisbane, Australia, in 1998 and 2005. Since mentation of VFDM-based opportunistic systems: issues 2006, he has been on the faculty of the Uni- and challenges,” REV Journal on Electronics and Commu- versity of Engineering and Technology (UET), nications, vol. 2, no. 1–2, pp. 1–18, 2012. Vietnam National University, Hanoi (VNU), [16] M. Maso, E. Baástu˘g, L. S. Cardoso, M. Debbah, and where he is currently an associate professor of ệ. ệzdemir, “Reconfigurable cognitive transceiver for electronic engineering in the Faculty of Elec- opportunistic networks,” EURASIP Journal on Advances tronics and Telecommunications and director in Signal Processing, vol. 2014, no. 1, pp. 1–18, 2014. of the Advanced Institute of Engineering and [17] L. S. Cardoso, M. Maso, and M. Debbah, “A green ap- Technology (AVITECH). He is interested in signal processing meth- proach to femtocells capacity improvement by recycling ods, including time-frequency signal analysis, blind processing, adap- tive filtering, compressive sampling, tensor-based signal analysis, wasted resources,” in IEEE Wireless Communications and graph signal processing, and apply them to wireless communication Networking Conference (WCNC), 2013, pp. 3817–3822. and networking, biomedical engineering. Nguyen Quoc Tuan was born in Hoa Binh, Tran Thi Thuy Quynh was born in 1979. Vietnam, in 1953. He received the B. Eng. She received the B.Sc., M.Sc. and Ph.D. de- degree in Radio Physics from University of grees in Telecommunication Engineering from Hanoi in 1980. He was a faculty member of the University of Engineering and Technology, Hanoi University of Science. Since 1996, he Vietnam National University, Hanoi, Vietnam has been on the Faculty of Electronics and in 2001, 2005, and 2016 respectively. Since Telecommunications, University of Engineer- 2009, she has been on the Faculty of Elec- ing and Technology, Vietnam National Univer- tronics and Telecommunications, VNU-UET sity, Hanoi, where he is currently an Associate as a researcher. Her research interests include Professor in Telecommunication Engineering. microwave component and antenna design, His research interests include radio physics, applying signal processing methods for an- communication and networking technologies, including signal pro- tenna arrays, and current focusing on implementing of test-beds for cessing and system programming. networking and cyber-security. Ngo Khac Hoang received the B.Eng. degree (Hons.) in electronics and telecommunications Ejder Bastug is a Member of Technical from University of Engineering and Technol- Staff at Bell Labs since 2018, associated ogy, Vietnam National University, Hanoi, Viet- with ENSA lab, coordinating with Sameer nam, in 2014; and the M.Sc. degree (Hons.) Sharma, concentrating on disruptive network- and Ph.D. degree in wireless communications ing paradigms as part of Future X vision. from CentraleSupộlec, Paris-Saclay University, Between 2016-2017, he was a postdoctoral re- France, in 2016 and 2020, respectively. His searcher at MIT and CentraleSupộlec, working Ph.D. thesis was also realized at Mathematical with Muriel Mộdard and Mộrouane Debbah and Algorithmic Sciences Laboratory, Huawei respectively. He obtained his Ph.D. at Cen- Technologies France. Since September 2020, he traleSupộlec in December 2015, focusing on has been working as a postdoctoral researcher at Communication distributed caching methods in small cell net- Systems Group, Department of Electrical Engineering, Chalmers works. He was in the executive or technical committee of flagship University of Technology, Gothenburg, Sweden. He is also an adjunct conferences, including IEEE WCNC 2014, EuCNC 2015, IEEE Black- researcher at the Advanced Institute of Engineering and Technology SeaCom 2015, IEEE ICC 2017, PIMRC 2019, and WCNC 2021. He is of VNU-UET. His research interests include wireless communications also a regular TPC member of IEEE conferences, and chair of several and information theory, with an emphasis on MIMO, noncoherent international workshops. He is currently an associate editor of IEEE communications, edge computing, massive random access, coded Communications Letters. His research interests are in the field of caching, and network coding. deterministic networking, machine learning and communications.
12. T. T. T. Quynh et al.: Network Coding with Multimedia Transmission and Cognitive Networking 83 Sylvain Azarian is currently Chief Technical Tran Xuan Nam is currently a professor and Officer of SDR-Technologies, a French startup head of the research group in advanced wire- working on distributed spectrum sensing and less communications in Le Quy Don Technical anomaly detections. He funded the company University, Vietnam. He received his Master in 2016 while he was project manager to of Engineering (ME) in Telecommunications the Director the Electromagnetism and Radar Engineering from University of Technology Departement (DEMR) at ONERA, the French Sydney, Australia in 1998, and Doctor of En- Aeronautics, Space and Defense research lab gineering in Electronic Engineering from The and Director of SONDRA, a joint research lab- University of Electro-Communications, Japan oratory between France and Singapore. From in 2003. From November 2003 to March 2006 2009 to 2014 he was Research Engineer at he was a research associate at the Information the Alcatel Lucent Chair on Flexible Radio. His research inter- and Communication Systems Group, Department of Information ests include wireless communications, signal processing, radar and and Communication Engineering, The University of Electro- Com- cognitive radio. munications, Tokyo, Japan. Dr. Tran’s research interests are in the areas of space-time signal processing for communications such as adaptive antennas, space-time coding, MIMO, spatial modulation and cooperative communications. Dr. Tran is a recipient of the 2003 IEEE AP-S Japan Chapter Young Engineer Award, and a co-recipient of two best papers from The 2012 International Conference on Advanced Technologies for Communications and The 2014 National Conference on Electronics, Communications and Information Technology. He is a member of IEEE, IEICE and the Radio-Electronics Association of Vietnam (REV). Le Vu Ha is currently Head of the Signals and Systems Laboratory at Faculty of Electronics and Telecommunications, VNU University of Engineering and Technology (VNU UET). He is also leader of the Neurotech research group of the Advanced Institute of Engineering and Technology (AVITECH) at VNU UET. His re- search and teaching interests include brain- computer interface, machine vision and 3D Mộrouane Debbah received the M.Sc. and vision, automatic image and video annotation, Ph.D. degrees from the Ecole Normale super-resolution imaging, bio-medical imag- Supộrieure Paris-Saclay, France. He was with ing and signal processing, and multimedia communications. Motorola Labs, Saclay, France, from 1999 to 2002, and also with the Vienna Research Cen- ter for Telecommunications, Vienna, Austria, until 2003. From 2003 to 2007, he was an Assistant Professor with the Mobile Com- munications Department, Institut Eurecom, Sophia Antipolis, France. In 2007, he was ap- pointed Full Professor at CentraleSupelec, Gif- sur-Yvette, France. From 2007 to 2014, he was the Director of the Alcatel-Lucent Chair on Flexible Radio. Since 2014, he has been Vice- President of the Huawei France Research Center. He is jointly the director of the Mathematical and Algorithmic Sciences Lab as well as the director of the Lagrange Mathematical and Computing Research Center. He has managed 8 EU projects and more than 24 national and international projects. His research interests lie in fundamental Vo Nguyen Quoc Bao is an associate pro- mathematics, algorithms, statistics, information, and communication fessor of Wireless Communications at Posts sciences research. and Telecommunications Institute of Technol- He is an IEEE Fellow, a WWRF Fellow, and a Membre ộmộrite SEE. ogy (PTIT), Vietnam. He is currently serv- He was a recipient of the ERC Grant MORE (Advanced Mathematical ing as the Dean of Faculty of Telecommu- Tools for Complex Network Engineering) from 2012 to 2017. He was nications and the Director of the Wireless a recipient of the Mario Boella Award in 2005, the IEEE Glavieux Communication Laboratory (WCOMM). His Prize Award in 2011, the Qualcomm Innovation Prize Award in research interests include wireless communi- 2012 and the 2019 IEEE Radio Communications Committee Technical cations and information theory with current Recognition Award. He received more than 20 best paper awards, emphasis on MIMO systems, cooperative and among which the 2007 IEEE GLOBECOM Best Paper Award, the cognitive communications, physical layer se- Wi-Opt 2009 Best Paper Award, the 2010 Newcom++ Best Paper curity, and energy harvesting. Award, the WUN CogCom Best Paper 2012 and 2013 Award, the He is the Technical Editor in Chief of REV Journal on Electronics 2014 WCNC Best Paper Award, the 2015 ICC Best Paper Award, the and Communications. He is also serving as an Associate Editor of 2015 IEEE Communications Society Leonard G. Abraham Prize, the EURASIP Journal on Wireless Communications and Networking, an 2015 IEEE Communications Society Fred W. Ellersick Prize, the 2016 Editor of Transactions on Emerging Telecommunications Technolo- IEEE Communications Society Best Tutorial Paper Award, the 2016 gies (Wiley ETT), and VNU Journal of Computer Science and Com- European Wireless Best Paper Award, the 2017 Eurasip Best Paper munication Engineering. He served as a Technical Program co-chair Award, the 2018 IEEE Marconi Prize Paper Award, the 2019 IEEE for ATC (2013, 2014, 2018), NAFOSTED-NICS (2014, 2015, 2016), REV- Communications Society Young Author Best Paper Award and the ECIT (2015, 2017), ComManTel (2014, 2015), and SigComTel (2017, Valuetools 2007, Valuetools 2008, CrownCom 2009, Valuetools 2012, 2018). He is a Member of the Executive Board of the Radio-Electronics SAM 2014, and 2017 IEEE Sweden VT-COM-IT Joint Chapter best Association of Vietnam (REV) and the Electronics Information and student paper awards. He is an Associate Editor-in-Chief of the jour- Communications Association Ho Chi Minh City (EIC). He is currently nal Random Matrix: Theory and Applications. He was an Associate serving as vice chair of the Vietnam National Foundation for Science Area Editor and Senior Area Editor of the IEEE TRANSACTIONS and Technology Development (NAFOSTED) scientific Committee in ON SIGNAL PROCESSING from 2011 to 2013 and from 2013 to Information Technology and Computer Science (2017-2019). 2014, respectively.
13. 84 REV Journal on Electronics and Communications, Vol. 10, No. 3–4, July–December, 2020 Pierre Duhamel received the Eng. Degree sur Yvette, France), where he developed studies in Signal processing in Electrical Engineering from the National for communications and signal/image processing for multimedia Institute for Applied Sciences (INSA) Rennes, applications, including source/protocol/channel coding/decoding. France in 1975, and the Dr. Eng. and the D.Sc He is also investigating the connections between communication degrees from Orsay University, Orsay, France theory and networking as well as information theory and AI. He in 1978 and 1986, respectively. has been “directeur de recherches ộmộrite” since March 2019. From 1975 to 1980, he was with Thomson- He has published more than 100 articles in international journals, CSF, Paris, France, where his research interests more than 300 papers in international conferences, and holds 29 included circuit theory and signal processing. patents. He is a co-author of the book “Joint Source and Channel In 1980, he joined the National Research Cen- Decoding: A cross layer perspective with applications in video broad- ter in Telecommunications (CNET), Issy les casting”; which appeared in 2009, Academic Press. He successfully Moulineaux, France, where his research activities were first concerned advised or co-advised more than 60 PhD students, and two of them with the design of recursive CCD filters. Later, he worked on fast are now fellows of the IEEE. algorithms for computing various signal processing functions (FFT’s, Dr. Duhamel is a fellow of EURASIP in 2008. He was awarded the convolutions, adaptive filtering, and wavelets. From 1993 to Sept. “grand prix France Telecom”; by the French Science Academy in 2000. 2000, he has been professor with ENST (National School of Engineer- He was a Distinguished lecturer, IEEE, in 1999, and was co-technical ing in Telecommunications), Paris with research activities focused chair of ICASSP 06, Toulouse, France and WCNC 2012, Paris, France. on Signal processing for Communications. He was the head of the A paper on subspace-based methods for blind equalization, which Signal and Image processing Department from 1997 to 2000. He is he co-authored, received the “Best paper award” from the IEEE currently with CNRS/LSS (Laboratoire de Signaux et Systemes, Gif Transactions on Signal Processing in 1998.