Linearization techniques for wideband, low-noise, CMOS software-defined radio receivers

Abstract

A communication revolution is happening with mobile communication equipment getting cheaper and feature rich, thanks to the continuing scaling of the digital CMOS process technology. In contrast, spectrum as a resource is becoming increasingly scarce, fragmented and crowded. Each country has its own spectrum allocation laid out for the various services making it challenging to design communication equipment that can operate across the globe. Especially, receiving weak wanted signals in the presence of strong out-of-band (OOB) interference which are quite close to the wanted signal band edge poses a challenging problem. This problem is traditionally mitigated by using dedicated and expensive high-quality band-pass filters in front of the receivers. In multi-standard mobile phones, these filters increase the cost and size of the phones. This is because these filters are not tunable, i.e. they work in a fixed frequency band, which makes them unusable if many radio standards in many different frequency bands need to be supported.Software-Defined Radio (SDR) receivers reduce the cost and form-factor of the receivers by using a re-configurable wide-band RF front-end while avoiding dedicated filters. However, this requires very linear front-end circuits and ways to implement the filtering on chip. To reject an OOB interferer, filtering requires separation in frequency of the interferer from the wanted signals. This was traditionally achieved by allocating guard bands between the frequency bands in which no emissions were allowed. However, to improve spectral efficiency these guard bands are made small or are completely eliminated. Furthermore, newer standards support wider bandwidths for higher data rates. In such a wideband receiver, even with some guard bands, the OOB interferers are so close to the wanted in-band signals that they are hardly filtered. Therefore, they are treated as effectively in-band signals. To improve the receiver's linearity, frequency translated filtering such as N-path techniques can be used to reject OOB interference but this is less effective to handle close-in OOB interference particularly for wideband radios.In this thesis, linearization techniques for wideband, low-noise, CMOS SDR receivers are presented for both in-band or close-in OOB as well as far OOB interference. Existing SDR receiver techniques are systematically reviewed to arrive at a receiver architecture using a frequency translated noise cancellation technique which is suitable for wideband and Low-Noise performance.In such a receiver, the RF transconductor which forms a part of the Low-Noise Transconductance Amplifier (LNTA) is the linearity bottle-neck for far OOB interference. A CMOS inverter as a transconductor can potentially achieve good linearity but relies on distortion cancellation making its linearity performance sensitive to process, voltage and temperature (PVT) variations. Therefore, a new PVT robust RF transconductor which uses resistive degeneration in combination with a ``floating battery'' technique that prevents second to third order distortion conversion is presented.A proof-of-concept RF transconductor implemented in a 45nm CMOS process while consuming 67% extra power, achieves an PIIP3 of 8dBm compared to an inverter with 2dBm PIIP3. The 1-dB compression was also improved by 5dB while the noise performance is only slightly degraded. When the chip was heated to 150oC the PIIP3 degrades by only 0.55dB proving its robustness to temperature.In the chosen receiver, a trans-impedance amplifier (TIA) is used as the first IF filtering stage to provide a low impedance to linearize a passive mixer and the RF transconductor driving the mixer. The TIA forms the linearity bottle-neck for close-in OOB interference provided that the LNTA is sufficiently linear. Therefore, a wideband distortion cancellation technique is proposed that cancels the distortion of the TIA's OPAMPs. Additionally, we present a calibration technique to achieve this distortion cancellation. Furthermore, a technique to improve the input matching of the receiver which is limited at higher frequencies by the input capacitance of the RF transconductor is presented.To verify the validity of the proposed techniques, a 2.5-4.5 dB NF, 100 MHz RF bandwidth, 0.5-2.5 GHz RF center frequency, zero-IF receiver was implemented in a 45 nm CMOS process. The linearization technique in combination with the calibration method when applied to the TIA improves the PIIP3 of the receiver to in-band and close-in OOB interference by 5-12 dB.This measured improvement in linearity was lower than the expectations from simulations. To investigate this, simulations were performed using simplified macro models of parts the receiver to verify the functioning of the proposed techniques. The time variant nature of the mixer, driving the TIAs was found to be the root cause of the worse-than-expected improvement in PIIP3. It also leads to the failure of the calibration method. Still, it was found that the proposed techniques work if the TIA is not interacting directly with a time variant circuit.