Circuits and architectures for broadband spectrum channelizers with sub-band gain control
Broadband receiver architectures for full-band or concurrent multi-band reception of signals are required in several applications. One approach to implementing such receivers is a spectrum channelizer that employs a frequency-folded analog-to-digital converter (FF-ADC). The design downconverts and channelizes a broadband input signal into multiple sub-bands at baseband by employing the harmonics of non-overlapping rectangular clocks. The downconverted and aliased baseband signal in each path is digitized by a baseband ADC, referred to as a sub-ADC below, that operates with a sampling rate that is lower than the Nyquist sampling rate set by the full bandwidth of the input signal. Sub-band separation is performed through digital harmonic rejection (HR) and image rejection (IR). The design operates similar to a time-interleaved ADC, except that it significantly reduces the bandwidth requirement of the samplers. If rectangular pulse waveforms are used in the FF-ADC down-converter, all sub-bands experience nearly equal gain during frequency down-conversion. Since all sub-bands are aliased to baseband before they are separated in the digital domain, a sub-band with large relative power can reduce the sub-ADC dynamic range that is available for other sub-bands, in addition to appearing as a blocker for other sub-bands. The research presented in this dissertation addresses approaches to overcome this issue, by embedding sub-band gain control within an FF-ADC. Chapter 2 proposes an approach that employs pulse-width-modulated local oscillator (PWM-LO) waveforms in the polyphase paths of an FF-ADC for scaling individual sub-band signal levels at baseband before digitization. The PWM-LO waveforms, which directly drive switches in each path, can be used to vary the gain in each sub-band by varying the level of harmonics in the waveforms. This is achieved by controlling the pulse-widths of the PWM-LO waveforms. This design avoids the requirement for N ×N switch matrices and variable transconductance cells in prior demonstrated approaches. The proposed architecture makes it possible to vary the spectral response of the FF-ADC with low signal-path complexity. Prediction of pulse widths for the desired harmonic, and hence the gain profile across all sub-bands, is performed using an off-chip supervised learning approach employing a neural network. Chapter 3 presents the implementation of a spectrum channelizer employing the PWM-LO-based sub-band amplitude control. The design allows for scaling the relative gain of the sub-bands over a 20-dB range. This relaxes the compression performance of the channelizer baseband and the sub-ADC dynamic range in the presence of sub-bands with significantly higher signal levels. Gain control on individual sub-bands is performed by employing customized PWM-LO waveforms,where the PWM-LO pulses are generated using delay-locked loops (DLLs). The off-chip neural-network based learning technique for estimating the PWM symbol pulse widths required for setting the desired LO harmonic levels is described. A 1.6 GS/s spectrum channelizer IC is implemented in a 65-nm CMOS process to verify the architecture. The measured channelizer gain is 51.6-56.5 dB without gain scaling and provides a range of 37-59 dB with PWM-LO gain control. Gain-scaling at a specific harmonic improves blocker compression in an unattenuated sub-band from -34 dBm to -16 dBm. The in-band gain compression with gain-scaling also increases from -32 dBm to -17 dBm. Chapter 4 describes a spectrum channelizer that uses voltage-mode downconversion. The approach requires a single voltage-mode input amplifier to drive the downconversion switches. Frequency-folding and sub-band gain control are achieved in a single signal path. This contrasts with the current-mode approach that requires a main FF-ADC path and a separate auxiliary path for sub-band gain control. By avoiding the requirement for an auxiliary input path, the approach presented here significantly simplifies the signal chain with identical gain-scaling capability. The contributions of this research and scope for future related work are summarized in Chapter 5.