Design and Implementation of Adaptive Multi-Tap Analog Interference Canceller
Abstract: Due to the severity of self-interferences, analog interference canceller is an essential part of a full duplex radio system. As the self-interference can potentially be even much stronger than the received signal, it must be meritoriously suppressed before analog-to-digital conversion. This paper is aimed to provide an optimal radio frequency domain multi-tap interference cancellation design. To this end, the following two key problems are addressed: the hardware design problem and the adaptive weight adjusting problem. For optimal hardware design, a tradeoff between widening effective delay range and maximizing the minimum cancellation capability is formulated. The principles and guidelines for optimal solution and hardware parameter design are derived. We also formulate adaptive weight adjusting as a multi-dimensional optimization problem, which is proved to be non-convex while adjusting the attenuators and phase shifters independently. By using a coordinate transformation technique, it is then converted into a convex optimization problem, and a convergent iterative algorithm based on the gradient descent method is applied to obtain the global optimum. The hardware implementation and experimental results substantiated the effectiveness of the proposed principles and algorithms.
Existing system: The RF domain analog SI canceller was first designed as a single-tap filter with a fixed or a programmable delay line for full-duplex systems . Although it proved promising for having low hardware and computational complexity, it has obvious drawbacks; most importantly, it can perfectly cancel the SI only when its delay is exactly equal to the tap’s delay. As a result, it requires tight component tolerances for precise programming which cannot be fully satisfied in practice. Another issue with single-tap analog SI canceller is that, in the absence of tightly coupled programmable delay, it can ideally cancel the SI only on a single frequency; thus leading to a significant performance loss in wideband communication systems. To solve the problems associated with single-tap analog canceller, multi-tap RF canceller was proposed, in which each tap of the FIR filter consists of a fixed delay line, a tunable attenuator and a tunable phase-shifter. Proposed system: Radio frequency (RF) domain analog SI cancellation is an effective way to suppress both the thermal noise from the transmitter’s amplifier and the nonlinear distortions caused by SI, and to prevent saturation of the analog-to-digital converters (ADCs). The following two types of RF SI cancellation schemes have been previously proposed. The first technique is based on reconstructing a baseband SI using SI channel estimates, and then converting it into RF domain for SI cancellation. However, this scheme is unable to cancel nonlinear distortions and phase noise introduced by the transmitter. To overcome this problem, in the second technique, the RF SI is directly reconstructed using the transmitted RF measurements. An SI canceller based on this technique consists of at least the following three parts: an analog finite impulse response (FIR) filter for reconstructing the interference using the reference signal fed from the transmitter, a combiner for subtracting the reconstructed SI from the received signal, and a weight adjusting unit for adaptively updating the weights of the FIR filter. Depending on the weight adjusting scheme, an analog SI canceller can be divided into two categories. Advantages:
Radio frequency (RF) domain analog SI cancellation is an effective way to suppress both the thermal noise from the transmitter’s amplifier and the nonlinear distortions caused by SI, and to prevent saturation of the analog-to-digital converters (ADCs). The following two types of RF SI cancellation schemes have been previously proposed. The first technique is based on reconstructing a baseband SI using SI channel estimates, and then converting it into RF domain for SI cancellation. However, this scheme is unable to cancel nonlinear distortions and phase noise introduced by the transmitter. To overcome this problem, in the second technique, the RF SI is directly reconstructed using the transmitted RF measurements. Disadvantages: As a result, it requires tight component tolerances for precise programming which cannot be fully satisfied in practice. Another issue with single-tap analog SI canceller is that, in the absence of tightly coupled programmable delay, it can ideally cancel the SI only on a single frequency; thus leading to a significant performance loss in wideband communication systems. To solve the problems associated with single-tap analog canceller, multi-tap RF canceller was proposed, in which each tap of the FIR filter consists of a fixed delay line, a tunable attenuator and a tunable phase-shifter. One of the key problems in multi-tap canceller design is to select appropriate hardware parameters. Modules: Self – interference: FULL duplex radios transmit and receive signals on the same frequency band simultaneously, thus providing not only a potential capability to double the spectral efficiency, but also the benefit of flexible and dynamic frequency resources allocation for uplink and downlink transmissions. This can be of particular importance in cellular systems for which it can be translated into a significantly improved network capacity. The main challenge in implementing full duplex systems, however, lies in the ability to suppress Self- Interference (SI) from the transmit chain to its own receive chain. With no propagation attenuation, the SI is much stronger than the desired received signal. In general, we require 90_110 dB
SI suppression to ensure that the interference is below the level of the receiver noise floor. Radio frequency (RF): Domain analog SI cancellation is an effective way to suppress both the thermal noise from the transmitter’s amplifier and the nonlinear distortions caused by SI, and to prevent saturation of the analog-to-digital converters (ADCs). The following two types of RF SI cancellation schemes have been previously proposed. The first technique is based on reconstructing a baseband SI using SI channel estimates, and then converting it into RF domain for SI cancellation. However, this scheme is unable to cancel nonlinear distortions and phase noise introduced by the transmitter. To overcome this problem, in the second technique, the RF SI is directly reconstructed using the transmitted RF measurements. An SI canceller based on this technique consists of at least the following three parts: an analog finite impulse response (FIR) filter for reconstructing the interference using the reference signal fed from the transmitter, a combiner for subtracting. Finite impulse response: An SI canceller based on this technique consists of at least the following three parts: an analog finite impulse response (FIR) filter for reconstructing the interference using the reference signal fed from the transmitter, a combiner for subtracting the reconstructed SI from the received signal, and a weight adjusting unit for adaptively updating the weights of the FIR filter. Depending on the weight adjusting scheme, an analog SI canceller can be divided into two categories. One is to adjust the weights based on the estimated SI channel state information (CSI); and the other is to use a recursive algorithm based on the observation of the canceller’s output. The former scheme requires individual observation and control chains on each tap, and accurate calibration of the analog chains. As such, in this paper, we have focused on the recursive weight adjustment algorithms using the canceller’s output observation. RF Domain multi – tap: In this paper, we addressed the RF domain multi-tap SI canceller design problem. Designing hardware taps and optimally adjusting their complex weights are the
main design challenges we faced. The residual interference power was theoretically and experimentally analyzed as a function of the system parameters including delay interval and delay range. We found that based on the Wiener solution for the weights, for some particular values of the delay interval, the canceller may encounter severe failure. The analysis identified invalid delay interval regions that should be strictly forbidden in system parameter design. Formulating a trade-off between widening delay range and maximizing the minimum cancellation capability, principles and guidelines for parameter design were also provided. Nevertheless, the main contribution of this paper is a novel solution to the non-convex optimization problem for weights calculation. By using a coordinate transformation method, the non-convex problem was converted into a convex one, and a gradient descent algorithm was developed. This leads to an iterative scheme with low implementation cost and fast convergence to the global optimum. Finally, we designed and implemented a hardware prototype achieving 38 dB SI cancellation gain, which substantiates the effectiveness of the proposed parameter design and weights adjustment algorithm.