tiadc-mismatch-model

KO | EN

gitlite — search

#typescript #ai-agents #ai #dsh-plugin #deepseek-harness #open-source #claude-code #codex #cli #developer-tools #react #windows

tiadc-mismatch-model

★ 28 Open GitHub ↗

Reproducible MATLAB behavioral model for TI-ADC mismatch generation, spectral validation, and calibration benchmarking.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# TIADC Mismatch Model

[![License: BSD-3-Clause](https://img.shields.io/badge/License-BSD--3--Clause-blue.svg)](https://raw.githubusercontent.com/SteveLuH/tiadc-mismatch-model/main/LICENSE)
[![MATLAB](https://img.shields.io/badge/MATLAB-R2022b%2B-orange.svg)](https://www.mathworks.com/products/matlab.html)
[![MATLAB tests](https://github.com/SteveLuH/tiadc-mismatch-model/actions/workflows/ci.yml/badge.svg)](https://github.com/SteveLuH/tiadc-mismatch-model/actions/workflows/ci.yml)

A reproducible MATLAB behavioral model of a time-interleaved analog-to-digital converter (TI-ADC) for developing, debugging, and benchmarking calibration algorithms.

> **Terminology and independence.** In this repository, **TI** means **time-interleaved**. This project is not affiliated with, sponsored by, or endorsed by Texas Instruments Incorporated. The default values are illustrative behavioral parameters, not measurements from a commercial ADC.

[中文说明](README.zh-CN.md) · [Model definition](docs/MODEL.md) · [Validation record](docs/VALIDATION.md) · [Calibration integration](docs/INTEGRATION.md)

## Why this repository exists

Calibration papers often use different mismatch generators, FFT conventions, and random seeds, which makes algorithm comparisons difficult. This project provides a small, inspectable plant model with deterministic mismatch instances and tests for the expected TI-ADC spectral signatures.

It intentionally contains **no calibration algorithm**. Your method receives the uncalibrated output and can be evaluated without inheriting a reference implementation's assumptions.

## Features

- Arbitrary channel count and sample rate.
- Per-channel offset, gain, static timing skew, first-order bandwidth mismatch, and second/third/fifth-order memoryless nonlinearity.
- Input-referred white noise followed by a signed mid-tread quantizer with exactly `2^B` codes.
- Reproducible local random streams that do not alter MATLAB's global RNG state.
- Coherent/noncoherent spectrum analysis, SFDR, SNDR, predicted mismatch image bins, and aggregate image energy.
- Optional illustrative temperature and normalized-supply drift.
- MATLAB unit tests and GitHub Actions CI; the core model requires only MATLAB.

## Quick start

```matlab
addpath('src');

cfg = tiadc.defaultConfig();
N = cfg.recordLength;
toneBin = 1001;
fin = toneBin * cfg.sampleRateHz / N;  % coherent tone

x = tiadc.makeTone(fin, -6, cfg, N);
mismatch = tiadc.generateMismatch(cfg, 42);
[y, state] = tiadc.simulate(x, cfg, mismatch);
metrics = tiadc.analyzeSpectrum(y, fin, cfg);

fprintf('SFDR %.2f dB, SNDR %.2f dB, image energy %.2f dBc\n', ...
    metrics.sfdrDb, metrics.sndrDb, metrics.imageEnergyDbc);
```

Run the complete example:

```matlab
run('examples/quickstart.m')
```

![Example uncalibrated spectrum](https://raw.githubusercontent.com/SteveLuH/tiadc-mismatch-model/main/docs/assets/quickstart-spectrum.png)

## Signal path

```mermaid
flowchart LR
    X["Periodic input record x[n]"] --> BW["Per-channel RC bandwidth"]
    BW --> SK["Static timing skew"]
    SK --> NL["a2/a3/a5 nonlinearity"]
    NL --> GO["Gain and offset"]
    GO --> NS["Input-referred noise"]
    NS --> Q["B-bit quantizer and clipping"]
    Q --> Y["Interleaved output y[n]"]
```

The front-end bandwidth and skew stages use an FFT-domain periodic steady-state model. See [the model definition](docs/MODEL.md) before using arbitrary bursts or transient waveforms.

## Public API

| Function | Purpose |
|---|---|
| `tiadc.defaultConfig` | Return the documented illustrative configuration. |
| `tiadc.generateMismatch` | Generate a deterministic channel-mismatch realization. |
| `tiadc.idealMismatch` | Return perfectly matched channel parameters. |
| `tiadc.simulate` | Run the behavioral converter model. |
| `tiadc.makeTone` | Generate coherent or noncoherent single/multitone records. |
| `tiadc.analyzeSpectrum` | Compute SFDR, SNDR, mismatch-image energy, and a spectrum. |
| `tiadc.mismatchImageBins` | Predict offset and gain/skew/bandwidth image locations. |
| `tiadc.generatePvtProfile` / `tiadc.applyPvt` | Generate and apply illustrative PVT drift. |

## Using your calibration algorithm

Start from [`examples/calibrator_template.m`](examples/calibrator_template.m). Treat the returned mismatch structure as hidden ground truth: it is useful for controlled oracle studies and debugging, but a blind/background calibration should not consume it.

A useful comparison should report at least:

- uncalibrated and calibrated SFDR/SNDR;
- the distribution over multiple held-out mismatch seeds;
- trained and untrained coherent tone bins;
- input amplitude and clipping rate;
- whether ground-truth mismatch parameters were used;
- MATLAB release and repository commit.

More detailed protocol guidance is in [docs/INTEGRATION.md](docs/INTEGRATION.md).

## Examples

- `quickstart.m` — one deterministic end-to-end run.
- `compare_mismatch_types.m` — isolated spectral signatures.
- `pvt_sweep.m` — illustrative temperature/supply sweep.
- `calibrator_template.m` — insertion point for an external algorithm.

## Validation and tests

```matlab
results = run_tests();
```

The tests check quantizer code count, ideal 12-bit SNDR, offset spurs at multiples of `fs/M`, the approximately linear-in-frequency timing-skew image amplitude, deterministic PVT drift, unique folded image bins, and RNG isolation.

The checked numerical values for release 0.1.0 are recorded in [docs/VALIDATION.md](docs/VALIDATION.md).

The repository is tested locally with MATLAB R2025b. CI is configured for R2022b and the latest MATLAB release using the official `matlab-actions` actions.

## Repository layout

```text
src/+tiadc/       Public MATLAB package
examples/         Runnable examples and integration template
tests/            matlab.unittest regression tests
docs/             Model equations, assumptions, and evaluation guidance
.github/          CI and contribution templates
```

## Scope and limitations

This is a behavioral research model, not a transistor-level, layout-extracted, or silicon-validated model. It currently assumes:

- periodic steady-state records for the FFT-domain analog front end;
- a single-pole channel bandwidth response;
- static channel timing skew rather than sample-to-sample aperture jitter;
- memoryless polynomial nonlinearity;
- independent illustrative mismatch/PVT distributions;
- no metastability, missing codes, clock phase noise, reference-network memory, package effects, or digital interface errors.

Do not claim measured hardware performance from results produced by this repository. Extend the plant model when an algorithm depends on an omitted mechanism, and document the extension in any comparison.

## Contributing, citation, and license

Bug reports, independent validation, new mismatch mechanisms, and calibration examples are welcome. See [CONTRIBUTING.md](CONTRIBUTING.md). Please do not contribute confidential code, proprietary PDK data, unpublished patent text, or data you do not have the right to redistribute.

Citation metadata is provided in [CITATION.cff](CITATION.cff). The project is licensed under the [BSD 3-Clause License](LICENSE).

tiadc-mismatch-model

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README