Advances in Inter-Channel Mismatch Calibration for Time-Interleaved Analog-to-Digital Converters: A Review

Abstract

Time-interleaved analog-to-digital converters (TIADCs) increase the aggregate sampling rate by distributing conversion across several lower-rate channels. However, differences in channel offset, gain, and sampling instant introduce periodic spurs and limit the dynamic performance obtainable from interleaving. This review examines the mechanisms of individual and combined mismatches and traces the development of calibration from foreground parameter identification to background adaptation, digital reconstruction, and all-digital joint compensation. Statistical estimation, correlation-based time-skew estimation, derivative and Taylor-series compensation, adaptive filtering, and fractional-delay and filter-bank reconstruction are compared in terms of reference requirements, operating assumptions, accuracy, and implementation cost. Particular attention is given to neural-network- and machine-learning-based calibration, including nonlinear mappings, temporal features, and hybrid representations that combine physical models with learned correction. These approaches can represent coupled errors, but their usefulness depends on training-data coverage, label accuracy, out-of-distribution generalization, and the cost of continuous inference. The discussion distinguishes background parameter adaptation from the continuous application of fixed coefficients, and separates measured-data validation from real-time hardware deployment. Future research should address broadband and complex-signal robustness, explicit channel information, lightweight fixed-point implementations, and stable online adaptation. Model–data hybrid methods are a promising direction when evaluated against appropriate conventional baselines. Reliable progress ultimately requires evidence from real converters and sustained data streams, with calibration accuracy assessed jointly with throughput, latency, resource use, and power.