Due to its potential to double the data rate at the Physical (PHY) layer and to provide many other benefits at the higher layers of the networking stack, full-duplex (FD) wireless has drawn significant attention [2][3][4][5] as an enabler of next-generation wireless networks. One of the major challenges associated with enabling FD wireless is the extremely strong self-interference (SI) on top of the desired signal. At transmit power levels greater than 10 dBm, more than 90 dB of SI cancellation (SIC) across the antenna interface, the RF/analog, and digital domains is required to sufficiently cancel the SI to below the radio's noise floor.
Within the Columbia FlexICoN project [6], we focus on the design of and experimentation with FD radios and systems grounded in integrated circuit (IC) implementations, which are suitable for hand-held and form-factor-constrained devices [7]. In [8], we presented the 1 stgeneration (Gen-1) narrowband FD radio and an FD link, featuring 40 dB RF SIC across 5 MHz bandwidth. This Gen-1 RF SI canceller emulates its RFIC counterpart presented in [9], whose modeling and analysis ✩ Preliminary version Kohli et al. (2020) of this paper was presented at the 14th International Workshop on Wireless Network Testbeds, Experimental Evaluation & Characterization (WiNTECH '20), Sept. 2020. [1].
* Corresponding author.
E-mail addresses: mpk2138@columbia.edu (M. Kohli), tingjun@ee.columbia.edu (T. Chen), mb4038@columbia.edu (M.B. Dastjerdi), jw3350@columbia.edu (J. Welles), seskar@winlab.rutgers.edu (I. Seskar), harish@ee.columbia.edu (H. Krishnaswamy), gil@ee.columbia.edu (G. Zussman).
were presented in [10]. In [11], we developed a 2 nd -generation (Gen-2) 19 wideband FD radio which leverages the technique of frequency-domain 20 equalization (FDE) to achieve over 50 dB RF SIC across 20 MHz band-21 width (4× improved bandwidth over to the Gen-1 canceller), which 22 emulates its RFIC counterpart presented in [12]. Both generations of 23 RF canceller achieve these levels of RF SIC while consuming less than 24 300 mW of power. 25 In order to allow the broader community to experiment with FD 26 wireless, we integrated the two generations of FD radios in the open-27 access ORBIT [13] and COSMOS [14,15] wireless testbeds. We provide 28 several variants of programmable FD radios, including those suitable 29 for static use and those suitable for mobile experimentation. Since inter-30 facing an RFIC canceller with a software-defined radio (SDR) presents 31 numerous technical challenges, we implemented the RF cancellers on 32 printed circuit boards (PCBs) to facilitate cross-layered experiments 33 with an SDR platform. Specifically, we integrated an improved version 34 of the Gen-1 RF canceller with a USRP N210 SDR in the ORBIT testbed [13]. In the COSMOS testbed [14,15], we integrated four Gen-1 2 RF cancellers (an improved version of that presented in [11]) with 2 USRP2 and USRP-2974 SDRs, one Gen-1 RF canceller with a USRP 3 N210 SDR, and a mobile FD radio based on the Gen-1 RF canceller. 4 In this paper, we first present the overall design of the Gen-1 and 5
Gen-2 FD radios. We then describe the various integrated FD radios 6 in ORBIT and COSMOS. Finally, we present three remotely accessible
Extensive reviews of research in the area of FD wireless were 29 presented in [2,3], including various implementations of FD systems, 30 and analysis of the influence of FD wireless at the higher layers of 31 the networking stack. Recent system-level design and evaluation of 32 FD systems has included mmWave FD systems [16], the use of FD 33 in wireless relay devices [17], and the deployment of long-range FD 34
backscatter tags [18]. 35 While [19][20][21] involved a pair of Tx and Rx antennas to achieve 36
Tx/Rx isolation at the antenna interface, we focus on FD radio designs 37 using a shared antenna interface such as a circulator, which are more 38 appropriate for single-antenna systems [4,22]. In addition, existing cir-39 cuit designs for RF/analog SIC often utilize a time-domain interpolation 40 approach utilizing parallel delay lines with amplitude and phase control 41 that are more suitable for discrete implementations on PCBs [4,23]. 42
We present two RF cancellers which utilize techniques suitable for 43 achieving RF SIC in compact IC implementations, while maintaining 44 the robust integration with an SDR that a PCB allows. 45
The development of FD radios and systems in previous research and 46 associated evaluation of PHY, medium access control (MAC) and higher 47 layer algorithms and techniques has largely taken place in laboratory testbeds or in simulation [16,17,22,[24][25][26][27][28]. The design and fabrication of FD hardware can itself be a daunting challenge, especially when networks of multiple FD devices are concerned. The integration of our FD hardware in ORBIT and COSMOS aims to alleviate this challenge. To the best of our knowledge, this paper presents the design and implementation of the first open-access, remotely accessible FD radios, together with opensource example experiments and FD software, that can be used by the community for experimentation with FD wireless.
The overall design of all FD radios integrated in ORBIT and COS-MOS, irrespective of generation or form factor, is shown in Fig. 1. The integrated FD radio design consists of three main components: an RF canceller box, a USRP SDR, and a remotely accessible compute node. Each Gen-1/Gen-2 canceller box has four components: a circulator, an antenna tuner, a SUB-20 controller, and an RF canceller PCB which differs between the two generations. Below, we briefly discuss these components. Coaxial Circulator. An RF-CI RFCR3204 coaxial circulator [29] is used, with operating frequency between 860 and 960 MHz. Programmable Antenna Tuner. To achieve better matching between the antenna and the circulator, we also designed and implemented a programmable antenna tuner at around 900 MHz frequency. Fig. 2 shows the circuit diagram and PCB implementation of the antenna tuner. In particular, a 𝜋-network with lossless inductor (𝐿) and digitally tunable capacitors (𝐶 𝑖 ) is used for impedance transformation. We use a fixed chip inductor with inductance 𝐿 f ixed = 5.1 nH and Peregrine Semiconductor PE64909 5-bit digitally tunable capacitors for 𝐶 𝑖 (𝑖 = 1, 2, 3). By programming capacitor 𝐶 𝑖 with code values CAPi, different antenna interface impedance matching can be achieved. The configuration ranges of the tunable capacitors are:
SUB-20 Controller. As Fig. 1 shows, a DIMAX SUB-20 multi-interface USB adaptor [30] connected to the remotely accessible compute node is used to program the tunable components on the RF canceller and antenna tuner through the serial peripheral interface (SPI). The SUB-20 SPI is configured to operate at the maximal SPI clock frequency of 8 MHz. The SUB-20 control for both the Gen-1 and Gen-2 canceller PCBs is handled through a customized GNU Radio out-of-tree (OOT) module (see Section 7.1 for details). RF Canceller PCB. In order to meet the USRP Rx front-end linearity and the analog-to-digital converter (ADC) dynamic range requirements, sufficient SIC in the RF domain is needed before digital SIC is performed. Therefore, the RF canceller PCB needs to provide up to 30 dB RF SIC in addition to the 20-25 dB provided by the circulator and antenna tuner. The RF canceller taps a reference signal from the output of the power amplifier (PA) at the Tx side, and SIC is performed at the input of the low-noise amplifier (LNA) at the Rx side. The difference between the Gen-1 and Gen-2 canceller boxes is the circuit design of the RF canceller PCB, as discussed in Sections 4.1 and 4.2 .
While the Gen-1 and Gen-2 FD radios share the same overall design described in Section 3, the core part of each radio, the canceller PCB, is different. In this section, we will describe the design of each generation of RF canceller PCB.
The Gen-1 RF canceller PCB is a narrowband frequency-flat amplitudeand phase-based canceller, which is an improved version of that presented in [8]. The RF canceller is implemented using discrete components on a PCB and is optimized around 900 MHz operating frequency. 1The reference signal is tapped through a 6 dB Mini-Circuits ADC-6-13+ directional coupler whose amplitude and phase are subsequently adjusted before SIC is performed at the Rx.
For amplitude adjustment, a SKY12343-364LF 7-bit digital attenuator is used, and for phase adjustment, a Mini-Circuits passive SPHSA-152+ phase-shifter is used. The phase shifter is controlled by a TI-DAC081S101 8-bit digital-to-analog converter (DAC). The attenuator and DAC have a 3 V supply voltage, and the phase shifter has a reference voltage of 12 V. The attenuator and DAC are programmed through the SUB-20 controller over SPI. The available parameter configuration ranges for the attenuator ATT (ATTenuation) and phase shifter PS (Phase Shift) are:
The attenuator and phase shift values can be input by the experimenter using the SUB-20 API via a graphical user interface (GUI)
The Gen-2 RF canceller box includes an FDE-based RF canceller 35 PCB which is an improved version of that described in [11]. 2 This 36 canceller PCB can achieve enhanced cancellation performance over a 37 significantly wider bandwidth compared to the Gen-1 canceller PCB, 38 and thus allows for experimentation with wider band signals. In this 39 section, we present the design and implementation of the Gen-2 FD 40 radio, including the Gen-2 RF canceller PCB. 41 The Gen-2 RF canceller illustrated in Fig. 5 is implemented using 42 discrete components on a PCB and is optimized around a 900 MHz op-43 erating frequency. 3 The reference signal coupled from the Tx is first fed 44 through an Analog Devices (ADI) HMC374 LNA before passing through
Summary of tunable components for the Gen-2 RF Canceller. 𝑖 ∈ {0, 1} represents the two FDE taps on the Gen-2 path.
Control Name Tuning range
an ADI HMC221B RF switch. This RF switch switches the reference signal between two paths: a Gen-1 narrowband path (identical to the Gen-1 RF canceller described in Section 4.1) and the wideband FDE path. In particular, the FDE path contains two parallel FDE taps, split and combined using Anaren PD0810J5050S2HF power dividers. Each FDE tap consists of a tunable bandpass filter (BPF) with amplitude and phase control.
Table 1 summarizes the tunable components on the Gen-2 canceller PCB. Each of the two paths contains an amplitude and phase control, identical to those on the Gen-1 canceller. Furthermore, each BPF is controlled by three tunable capacitors, one to control the center frequency of the BPF, and two identically controlled capacitors to control the quality factor of the BPF. For controlling the center frequency, we use the Peregrine Semiconductor PE64102 4-bit digitally tunable capacitor with a tuning range of 1.88-14.0 pF. For controlling the quality factor, we use Peregrine Semiconductor PE64909 5-bit digitally tunable capacitors with a tuning range of 0.6-2.35 pF. Together, these components provide a total of 2 48 possible configurations for the Gen-2 path. The Gen-1 path's amplitude and phase control can be considered fully independent of the FDE path, as the RF switch provides up to 30 dB of isolation between paths. Similarly to the Gen-1 canceller PCB, the experimenter can control the values in Table 1 via the GUI described in Section 7.1. Fig. 6 shows the effect of varying each parameter on the frequency response of one individual FDE tap. In each case, only one parameter is varied, while the others are kept at their lowest values. Fig. 7 shows the achievable RF SIC of the wideband FDE path, where >50 dB RF SIC is achieved across 20 MHz bandwidth. This is 5 dB higher RF SIC achieved over 2-4× the bandwidth when compared to the Gen-1 path, and means that higher bandwidth signals, such as orthogonal frequency-division multiplexing (OFDM) signals, may be used in experimentation.
Recall the goal of making the FD radios openly accessible to remote 33 experimenters. An ORBIT node equipped with the Gen-1 RF canceller 34 box is depicted in Fig. 9. We use node11-10 in the ORBIT main grid Gen-1 FD radio, experimenters may also utilize other nodes in the 1 testbed to create a wireless link where one radio operates in FD mode. 2
The USRP has a receiver noise floor of -86 dBm at 10 MHz bandwidth. 4 3 We developed a node image running Ubuntu (flexicon-orbit-4 v4.ndz), which contains the two example GNU Radio FD experiments 5 described in Sections 7.1 and 7.2 as well as GNU Radio and UHD 6 preinstalled [31,32]. While there is only one FD radio integrated in 7
ORBIT, the large number of controllable half-duplex (HD) radios can be 8 used to replicate many different experimentation scenarios on the node-9 level. For example, the node-level SIC performance can be investigated 10 as described in Section 7.1. Moreover, the performance in the presence 11 of many interferers can be studied. 12
In COSMOS Sandbox 2, located at Columbia University, we inte- 4 This USRP receiver noise floor is limited by the existence of environmental interference at 900 MHz frequency. The USRP has a true noise floor of around -95 dBm at the same receiver gain setting, when not connected to an antenna. is connected to the remotely accessible compute node over Ethernet, and the SUB-20 controller is connected over USB. The USRP2 SDR supports up to 25 MHz bandwidth over the 1 Gbps Ethernet connection to the remotely accessible compute node. Together, this configuration provides real-time FD operation up to 5 MHz bandwidth, the maximum operating bandwidth of the Gen-1 RF canceller box. The Gen-1 RF canceller box and its integration in COSMOS is shown in Fig. 12(a). While the Gen-2 FD radios all contain a Gen-1 path on the RF canceller PCB, the Gen-1 FD radio provides experimenters with simpler hardware with only two tunable parameters (amplitude and phase) as described in Section 4.1. As a result, the Gen-1 FD radio is easier to control and can provide sufficient RF SIC in narrowband use cases.
As illustrated in Fig. 10, two of the Gen-2 FD radios are connected to USRP2 SDRs, which are synchronized over a MIMO cable. allowing them to be used in a synchronized FD link. These two Gen-2 FD radios are shown in Fig. 12(b) with a 5 ft distance between antennas. The same remotely accessible compute node that controls the Gen-1 FD radio is also used to control these two Gen-2 FD radios. The higher bandwidth supported by the Gen-2 FD radio allows for running realtime FD experiments at 10 MHz bandwidth with a GUI. With the GUI disabled and when using the command line terminal, the supported bandwidth can be increased to the full 20 MHz supported by the Gen-2 RF canceller box.
Two more identical Gen-2 FD radios were also integrated in COS-MOS, which are connected to two USRP-2974 SDRs, synchronized using an OctoClock-G CDA-2990. The USRP-2974 SDR contains a built-in PC, and thus serves both as the SDR and the remotely accessible compute node. This PC has an Intel i7 2.0 GHz quad-core processor, 16 GB RAM and runs Ubuntu, GNU Radio and UHD. The two FD radios connected to the USRP-2974s can also be used as an FD link, similar to the two Gen-2 FD radios connected to USRP2 SDRs.
The inclusion of four wideband FD radios in Sandbox 2 is an important step in the development and experimentation of networklevel algorithms for FD and heterogeneous FD/HD networks. While two radios are sufficient for simple link-level experimentation, involving a network of more nodes facilitates investigation of the impacts of FD on the higher layers of the network stack (e.g. [33,34]) that is backed up by the use of real FD hardware. The four integrated Gen-2 FD radios all support similar RF SIC performance around 925 MHz, shown in Fig. 7, meaning experiments involving multiple FD radios can be well controlled.
The available Gen-1 and Gen-2 FD radios enable a wide vari-2 ety of FD experiments, and are statically placed in order for remote 3 experimenters to be able to conduct repeatable experiments. As exper-4 imentation with mobile user equipment (UE) operating in FD mode 5 is of interest (e.g. [35,36]), we have fabricated a prototype mobile 6
FD radio consisting of a Gen-1 RF canceller box similar to that in 7 Fig. 12(a), a USRP B205mini-i SDR, and an Intel NUC 8i7BEH. The 8 complete prototype mobile FD radio is shown in Fig. 13, and the RF SIC 9
performance of the integrated Gen-1 RF canceller is shown in Fig. 4. 10
The USRP B205mini-i is embedded within the RF box, leading to a 11 portable FD UE that can be powered with portable batteries. 12 For remote-controlled mobility, we will utilize an X-Y table whose 13 setpoint can be configured remotely through the COSMOS Sandbox 2 14 console (see [37] for such an X-Y
Recall that a core feature of the integrated FD radios is that they 30 can be remotely accessed by experimenters. The process for remotely 31 accessing the FD radios is the same for both ORBIT and COSMOS: the 32 experimenter logs into the testbed console from a local machine using 33 Secure Shell (SSH). X11 forwarding is used to enable the GNU Radio 34 The developed example experiments run in real-time, where the experimenter can observe results visualized without offline processing, and employ an orthogonal OFDM PHY layer with varying bandwidth and modulation and coding schemes (MCS). These example experiments, described below, can be used to benchmark several performance metrics of the Gen-1 and Gen-2 radios integrated in ORBIT and COSMOS, including node-level SIC and link-level PRR.
Real-time performance is achieved by the use of customized C++ out-of-tree (OOT) blocks. The implemented open-source OOT blocks are available at [39] and described in Section 7.1 below, and are also used in the experiments described in Sections 7.2 and 7.3 . Experimenters may use GNU Radio to develop their own FD experiments, either by leveraging the available OOT blocks or creating their own. For further improved performance, the experimenters can also develop experiments by directly invoking the UHD driver in C++ code.
In this experiment, the experimenter can observe visualized realtime performance of a single FD radio while transmitting OFDM packets. The following data is visualized: (i) the time domain Rx signal after RF SIC and after digital SIC, (ii) the power spectrum of the Rx signal after RF SIC, (iii) the power spectrum of the Rx signal after both RF and digital SIC, and (iv) the digital SIC filter taps. The results were achieved on the Gen-1 FD radio in ORBIT, and two Gen-2 FD radios in COSMOS, one connected to a USRP2 and the other to a USRP 2974. This experiment makes use of many customized OOT blocks, which are described below.
A core part of this experiment is the ability of the experimenter to manually configure the canceller PCB to achieve different RF SIC profiles. The experimenter has access to a GUI with controls that allow the full range of input described in Table 1. The numerical values set by these GUI controls are input to the SUB-20 control OOT block, which configures the canceller box using the SUB-20 application programming interface (API) [30] over SPI. For the Gen-1 and Gen-2 RF canceller boxes, a total of 7 and 23 bytes, respectively, are sent over SPI to program the canceller PCB and antenna tuner. With an 8 MHz SPI clock, one configuration of the Gen-1 and Gen-2 canceller PCBs takes 7 μs and 23 μs, respectively. Therefore, experimenters can observe different RF SIC profiles in real time.
In the stable indoor environment provided by COSMOS Sandbox 2 and the ORBIT grid, a given configuration can reliably provide the same SIC performance across different experiment runs. A C++based implementation of the adaptive canceller configuration method described in [11] is also considered as a subject of our future work. then be successfully decoded in real time using a receiver from the 9 aforementioned 802.11 framework. Together, the use of this framework 10 in this method allows for real-time transmission of realistic 802.11 data packets.
Packet Encapsulation. The implemented digital SIC algorithm requires additional pilot symbols prepended to every packet. We implemented a customized OOT block to prepend these symbols, and to add zero padding between packets.
Digital SIC is performed on the packet-level using a leastsquares algorithm to estimate the SI channel in the time domain with additional pilot OFDM symbols. The OOT block performs both the least-squares SI channel estimation and digital SIC with the following computation: 𝐫 Rx = 𝐲 Rx -𝐀 ̂𝐡. Here, 𝐫 Rx is the residual SI after RF and Fig. 15. RF SIC achieved over 16.7 MHz using WiFi-like OFDM packets for the Gen-1 FD radio integrated in ORBIT and Gen-2 FD radios integrated with USRP2 SDRs in COSMOS. The Gen-2 wideband FD radio can achieve a greater RF SIC over a wider bandwidth.
digital SIC, 𝐲 Rx is the signal after RF SIC, 𝐀 is the Toeplitz matrix 1 constructed from the transmitted signal, and ̂𝐡 is the estimated SI 2 channel (i.e., the digital SIC filter taps). 3 SIC Performance. Fig. 14 shows the node-level SIC performance of the 4
Gen-1 narrowband FD radio in ORBIT and the Gen-2 wideband FD radio 5 using the USRP2s and USRP 2974 in COSMOS. By configuring the RF 6 cancellers appropriately, the Gen-1 FD radio can achieve 40-45 dB RF 7 SIC across 3-5 MHz, and the Gen-2 FD radios can achieve 45-50 dB RF 8 SIC across 16.7-20 MHz. In both cases, the RF SIC is followed by 34-9 37 dB of digital SIC, resulting in an overall SIC of over 80 dB. Given the 10 low transmit power used in these experiments, 80 dB of overall SIC is 11 sufficient for successful FD operation. 12 For higher transmit power levels, an overall SIC of closer to 90 dB 13 is required for cancellation to the noise floor. The primary limitation 14 of the Gen-1 and Gen-2 FD radio SIC performance is the nonlinear 15 performance of the RF chain, most predominantly due the power 16 amplifier of the USRP and the RF canceller itself. These nonlinearities 17 manifest themselves as imperfect SIC at a level that is, in our case, 4-18 5 dB above the noise floor. When a higher transmit power is used, this 19 excess above the noise floor increases further. 20 A comparison of the RF SIC performance for the integrated Gen-1 21 and Gen-2 radios is illustrated over a wider bandwidth in Fig. 15, which 22 clearly shows the narrowband and wideband performances. Across the 23 16.6 MHz bandwidth shown in the figure, the Gen-1 FD radio can 24 achieve 30 dB RF SIC, whereas the Gen-2 FD radio can achieve 50 dB 25 RF SIC. 26
In this experiment, the experimenter can use the two Gen-2 FD 28 radios connected to USRP2 SDRs to perform link-level experimentation. 29
The link PRR as a function of the signal-to-noise ratio (SNR) is mea-30 sured while the two Gen-2 FD radios operate in HD or FD mode. To 31 achieve a successful FD link, the experimenter can configure each FD 32 radio via the methods described in Section 7.1 to achieve the desired 33 SIC across the RF and digital domains. The Tx power of the FD radios 34 is swept during the experiment. For each Tx power value, the FD 35 radios transmit 1,000 WiFi-like 100-byte data packets. The experiment 36 is repeated for all 8 MCS described in Section 7.1. 37
We consider two performance metrics. The first is the HD or FD link 38 SNR, which is measured as the ratio between the average Rx signal 39 power level (across all 1,000 packets) and the Rx noise floor when the 40 link operates in HD or FD mode. The second metric is the HD or FD 41 link PRR, which is calculated for each FD radio as the percentage of the 42 1,000 transmitted packets that are successfully received and decoded. Fig. 16 plots the measured HD and FD link PRR as a function of the HD link SNR. The results show that with sufficient link SNR values, both Gen-2 FD radios in a link can achieve a PRR of 1 while operating in FD mode. This corresponds to a link-level FD rate gain of exactly 2×. With insufficient link SNR values, there is a roughly 2 dB degradation in achieved SNR when operating in FD mode, caused by imperfect SIC, which causes a residual SI that is 1-2 dB above the noise floor. This leads to a reduction in PRR by 30%-50%, leading to an FD rate gain of 1.0-1.4×.
In particular, with an HD link SNR of 20 dB, the PRR for 64-QAM 3/4 in HD mode is 0.45, and the PRR for 64-QAM 2/3 in FD mode is 0.35. The data rate supported by the 802.11a 64-QAM 3/4 packet transmissions used in our experiments is 54 Mbps, and 48 Mbps for 64-QAM 2/3. Therefore, using the empirical PRR values from the results in Fig. 16, we can compute an effective data rate for 64-QAM 3/4 in HD mode of 24.3 Mbps. For 64-QAM 2/3 in FD mode, the computation is similar, except for an additional factor of 2 as the radios operate in FD mode, giving an effective data rate of 33.6 Mbps. This represents a data rate gain in FD mode of roughly 1.4×. If we perform a similar comparison at 20 dB SNR but with 64-QAM 3/4 also used for FD mode, we find a data rate reduction of almost 3×. This suggests that in insufficient SNR regimes, FD link gain may be achieved using a lower data rate MCS.
The experimenter can make use of multiple nodes (in HD or FD mode) for network-level experimentation. In Section 7.2, a link-level FD rate gain of up to 2× is demonstrated with sufficient link SNR values, and it is possible to investigate the impact of this result in a network of multiple FD radios. To demonstrate this capability, a simple slotted ALOHA-based protocol is used for medium access. Up to four Gen-2 FD radios can be used in such a network. Fig. 10 shows these four radios within Sandbox 2, while Fig. 17 shows the physical topology of the four radios, and the SNR between each radio pair is shown in Table 2 for -10 dBm transmit power and 5 MHz bandwidth at 915 MHz carrier frequency. The USRP 2974 and two USRP2s that are connected to the four Gen-2 FD radios are synchronized using an external clock source, allowing for the time-slotted operation.
In this experiment, we utilize three Gen-2 FD radios out of the available four shown in Fig. 17: radios #1 and #2 connected to the USRP-2974 and radio #3 connected to a USRP2. In each time slot with a duration of 22 ms, a radio may only send one packet with an over-the-air transmission time of 2 ms. For each radio, the average node throughput 𝑆 𝑖 , 𝑖 ∈ {1, 2, 3}, is defined as the ratio between the number of packets successfully decoded by radio 𝑖, 𝑛 𝑖 , divided by the total number of time slots 𝑇 : 𝑆 𝑖 = 𝑛 𝑖 𝑇 . 5 Similarly to Section 7.2, the experiment is run in two modes: (i) HD mode: all radios operate in HD and utilize a slotted ALOHA protocol with an equal transmission probability of 1 3 , and (ii) FD mode: all radios operate in FD and utilize a slotted ALOHA protocol with an equal transmission probability of 1 3 . The latter mode can result in scenarios where the network can now support two simultaneous transmissions that can each be correctly decoded at the receiver. In particular, a radio 𝑖 can successfully decode a packet in a given time slot if 𝑖 transmits and one other radio transmits.
Each experiment run consists of 20,000 time slots, during which each radio sends WiFi-like 100-byte data packets with BPSK 1/2 MCS. The average node throughput for each radio is then computed after the experiment completes. The SNR values between each radio pair during an ongoing transmission, a control mechanism traditionally limited to wired networks. Experimenters may use the synchronized Gen-2 radios to investigate such impacts on various network topologies, including the broadcast network used in the ALOHA experiment, as well as centralized networks where one node acts as a base station or access point. The integrated FD radios can also support experimentation with heterogeneous HD/FD networks, where relative priority of HD and FD nodes can impact network throughput and fairness [42]. In such experiments, especially those using a carrier sense-based MAC protocol [33], an important factor is the host-to-SDR latency, the optimization of which is part of our future work on additional customized C++ FD software.
In this paper, we presented our cross-layered (hardware and software) design and implementation of the first open-access, remotely accessible FD radios integrated in the ORBIT and COSMOS wireless testbeds. We make openly accessible a variety of FD radios based upon wideband and narrowband RF SI canceller designs, including fixed radios connected to powerful remotely accessible compute nodes, and a mobile radio connected to a portable PC. The presented example experiments along with the tutorial and open-source software [39,40] can be used on any integrated FD radio and expanded to different network scenarios. Therefore, we anticipate that the integrated FD radios and example experiments can facilitate further hands-on research in FD wireless.
In future work, we plan to make several improvements to the integrated FD radios. In particular, we will fully replace the USRP2s with USRP 2974s, which have superior performance that includes supporting a higher signal bandwidth and better RF chain linearity. This will facilitate a higher level of overall achievable SIC and the use of the 802.11a PHY layer waveforms [41], as well as support RF SI canceller 1 designs of higher bandwidth. We will also make the radios accessible 2 to the servers in COSMOS Sandbox 2, which provide heterogeneous 3 computing resources, including CPUs, GPUs, and FPGAs, as well as 4 configurable connectivity to COSMOS' optical network. Lastly, we will 5 make 28 GHz millimeter-wave (mmWave) phased array antenna mod-6 ules (PAAMs) [43] connected to the USRP N310 SDRs available to 7 researchers. These 28 GHz PAAMs will support research into FD in 8 mmWave bands [16] and using phased arrays [44]. 9
We anticipate that these improvements will allow for more complex
In this implementation, we select operating frequencies around the 902-928 MHz ISM band. However, the implementation can be easily extended to other frequencies (e.g., the
2.45 GHz and 5.8 GHz ISM bands).
The detailed design and evaluation of this canceller can be found in[11].
As before, the design can be modified to other sub-6 GHz frequencies.
Notice that in order to avoid the need for acknowledgment packets, the throughput is computed based on the number of received packets. While this is not the ''standard'' definition, it is easy to analytically evaluate such a small network.