Advanced Architectures for Self- Interference Cancellation in Full-Duplex Radios: Algorithms and Measurements Dani Korpi, Mona AghababaeeTafreshi, Mauno Piililä, Lauri Anttila, Mikko Valkama Department of Electronics and Communications Engineering, Tampere University of Technology, Finland
Outline Introduction How to achieve real-time inband full-duplex operation? Real-time nonlinear digital canceller Measurement results Conclusion 8.11.2016 2
Introduction Efficient self-interference (SI) cancellation methods have been widely studied in literature Various solutions have been demonstrated to work well in real-life conditions Hence, the principles needed for sufficiently high SI cancellation are more or less known The next step is to study how the SI cancellation can be achieved in real-time 8.11.2016 3
How to achieve real-time inband full-duplex operation? All the different cancellation stages must be capable of real-time processing Passive techniques very favourable in this respect For the active cancellation stages, simple algorithms and learning rules are needed RF domain cancellation Digital cancellation with nonlinear processing 8.11.2016 4
The considered full-duplex device architecture We consider a single-antenna device, where the TX and RX chains are separated by a circulator 20-30 db of passive isolation On top of that, two active SI cancellation stages are needed RF cancellation Digital cancellation (with nonlinear modeling) The active cancellation stages must be capable of real-time operation PA TX Original transmit data RF canceller NI 5791 RF transceiver Digital canceller RX Cancelled signal 8.11.2016 5
RF cancellation in real-time We have previously implemented a three-tap RF canceller with a real-time digital control algorithm 1 There, a simple LMS learning rule was implemented for controlling each tap weight Good cancellation performance in real-time 1 J. Tamminen, M. Turunen, D. Korpi, T. Huusari, Y.-S. Choi, S. Talwar, and M. Valkama, "Digitally-controlled RF self-interference canceller for full-duplex radios," in Proc. EUSIPCO, Aug. 2016, pp. 783-787. 8.11.2016 6
RF cancellation in real-time Within the digital weight calculation block, the weight of the ith tap is updated as w i k + 1 = w i k + μx i k e k μ is the learning parameter x i k is the ith tap signal e k is the cancelled signal Very simple learning algorithm, but the actual cancellation is still done in the RF domain in realtime TX to circulator From circulator τ n τ n n th Tap τ n Downconverter n th Tap Downconverter n th Tap Downconverter VectorModulator Vector Modulator + Vector Modulator Σ + + Q Q Q I I + Σ Weight Weight Calculation I Weight Calculation Calculation Feedback Downconverter From PA RX 8.11.2016 7
Digital cancellation in realtime Since the RF canceller cannot fully suppress the SI, digital cancellation is typically needed Furthermore, nonlinear modeling in the digital cancellation stage is necessary Learning the coefficients with a simple LMS algorithm ensures that the nonlinear cancellation can be performed in real-time 8.11.2016 8
Digital cancellation in realtime The cancellation and learning procedure is performed for each iteration as follows: y DC n = y RF n h n H u n h n + 1 = h n + μy DC n u n Here, the vector u n contains all the nonlinear basis functions and their delayed versions to model the memory effects φ p x n = x n p 1 x n Nonlinear SI estimate h n contains the current coefficient estimates 8.11.2016 9
Digital canceller implementation The nonlinear realtime digital canceller was implemented on an FPGA The basis functions were computed and orthogonalized offline Orthogonalization needed for LMS Original transmit data (x(n)) PA TX Pre-computed NI PXIe-7972R with Kintex-7 FPGA NI 5791 RF transceiver LPF L x(n) x(n) 2 x(n) x(n) P-1 x(n) RF canceller RX Orthogonalization h 1 h 2 h P Σ Σ Σ LMS filter weight update Cancelled signal 8.11.2016 10
Digital canceller implementation The digital canceller implementation utilizes 4 nonlinear basis functions With the memory taps, it amounts to 108 coefficients The coefficient update is simplified by forcing the step size to be a negative power of two Implemented as a bit shift Parameter Value Clock frequency of the canceller 130 MHz Sampling frequency of the signal 26 MHz Highest nonlinearity order 7 Memory length 27 Total number of coefficients 108 Step size 2-13 Delay of the canceller 17 cycles Number of bits at different stages Received signal 16 bits Basis functions (in memory) 25 bits Coefficients 25 bits Canceller output 16 bits 8.11.2016 11
Measurement results The performance of the real-time cancellation solutions is then evaluated with RF measurements The setup includes both RF and digital cancellation In some of the results, the RF canceller is replaced by attenuators This is done to accurately control the analog SI suppression Parameter Signal bandwidth Center frequency PA gain Receiver noise floor Value 20 MHz 2.45 GHz 24 db -88 dbm 8.11.2016 12
Calculating the true selfinterference power In the measurements, the same transmit signal sequence is transmitted repeatedly This allows for calculating the true power of SI over a certain time period In particular, the actual residual SI is a function of the transmit signal, and hence it also repeats Averaging over the repetitions removes noise, which is different for each repetition This can only be done over relatively short time periods to ensure that the channel conditions remain constant 2 z i n ~N 0, σ noise 1 L L i=1 y i,si n + z i n L = 1 L i=1 L y i,si n + 1 L i=1 z i n y SI n One repetition of the residual SI y i,si n y SI n 0 8.11.2016 13
Measurements: Different analog attenuation levels For the purposes of this figure, the RF canceller has been replaced by different attenuators Provides some initial idea about the performance boundaries of the real-time digital canceller The residual SI is neglibily low when the SI power entering the digital domain is -55 dbm or lower 8.11.2016 14
Measurements: Signal spectra at different stages With the actual RF canceller, this is achieved with a transmit power of roughly 7 dbm As can be observed, the residual SI is not quite as low as with an ideal attenuator, but it is still below the RX noise floor This corresponds to a case where both the RF and digital cancellers are tracking and cancelling the SI in real-time 8.11.2016 15
Measurements: Residual SI with respect to transmit power Here, the overall cancellation performance is evaluated for different transmit powers Active RF and digital cancellation, both in real-time The residual SI is above the noise floor only with the very highest transmit power Our current setup is limited to a transmit power of 12 dbm The real-time cancellation performance seems to be sufficient for TX powers up to 10 dbm 8.11.2016 16
Measurement: Tracking performance Here, the magnitudes of 5 most significant taps are shown, alongside with the total residual power The channel environment was purposefully disturbed during the measurement It can be seen that the weights are adjusted accordingly, while the total residual power remains constant Transmit power: 5 dbm RF canceller replaced with a 40 db attenuator 8.11.2016 17
Example application for the digital canceller Recently, we performed some measurements with a relay antenna, where digital cancellation was the only active cancellation stage 2 RF cancellation stage could be omitted due to the high isolation provided by the relay antenna (60-70 db) Digital cancellation was performed offline, but using a similar algorithm as here TX PA Original transmit data Back-to-back relay antenna NI PXIe-5645R Vector Signal Transceiver Digital canceller RX Cancelled signal 2 D. Korpi, M. Heino, C. Icheln, K. Haneda, and M. Valkama, "Compact inband full-duplex relays with beyond 100 db self-interference suppression: Enabling techniques and field measurements," IEEE Transactions on Antennas and Propagation, 2016, in press (available: tut.fi/full-duplex) 8.11.2016 18
Example application for the digital canceller Very good overall cancellation performance with only digital cancellation and 80 MHz bandwidth For a relay application, having only a nonlinear digital canceller could be enough! 8.11.2016 19
Conclusion In this work, we demonstrate a fully real-time and adaptive full-duplex device with two active cancellation stages The proposed architecture could cancel the SI quite efficiently as demonstrated by reallife measurements As future work items, we aim to generate the basis functions also in real time, while increasing the transmit power 8.11.2016 20
Thank you Questions? 8.11.2016 21