DTC Linearization via Mismatch-Noise Cancellation for Digital Fractional-N PLLs

Digital-to-time converter (DTC) based quantization noise cancellation (QNC) has recently been shown to enable excellent fractional- $N$ PLL performance, but it requires a highly-linear DTC. Known DTC linearization strategies include analog-domain techniques which involve performance tradeoffs and digital predistortion techniques which converge slowly relative to typical required PLL settling times. Alternatively, a DTC implemented as a cascade of 1-bit DTC stages can be made highly linear without special techniques, but such DTCs typically introduce excessive error from component mismatches which has so far hindered their use in low-jitter PLLs. This paper presents a background calibration technique that addresses this issue by adaptively canceling error from DTC component mismatches. The technique is entirely digital, is compatible with a large class of digital fractional- $N$ PLLs, and has at least an order of magnitude lower convergence time than the above-mentioned predistortion techniques. The paper presents a rigorous theoretical analysis closely supported by simulation results which quantifies the calibration technique’s convergence time and noise performance.

I. Introduction

The signal processing performed within any fractional- phase-locked loop (PLL) for frequency synthesis inevitably involves quantization. The resulting quantization error degrades the PLL’s phase noise unless it is actively canceled prior to frequency modulation, a process known as quantization noise cancellation (QNC). An increasingly popular QNC method uses a digital-to-time converter (DTC) to cancel most of the quantization error prior to phase error measurement within the PLL. This prevents the quantization noise from being subjected to the inadvertent but inevitable nonlinearity of the phase error measurement circuitry, thereby avoiding fractional spurs which would otherwise be caused by nonlinearly distorting the quantization error [1], [2], [3], [4], [5].