GitHub

Acquisition.jl

GNSS signal acquisition using the FM-DBZP (Frequency-domain Modified Double Block Zero Padding) algorithm.

Acquisition.jl is part of the JuliaGNSS ecosystem and works with GNSSSignals.jl for signal definitions.

Installation

using Pkg
Pkg.add("Acquisition")

Quick Start

using Acquisition, GNSSSignals
import Unitful: Hz

system = GPSL1CA()
(; signal, sampling_freq, interm_freq) = generate_test_signal(system, 1)

# Acquire GPS L1 C/A signals for PRNs 1–32
results = acquire(system, signal, sampling_freq, 1:32; interm_freq)

# Reuse a pre-computed plan for repeated acquisitions
plan = plan_acquire(system, sampling_freq, 1:32)
results = acquire!(plan, signal, 1:32; interm_freq)
Pick a fast sampling frequency

Acquisition performance depends strongly on the prime factorization of samples_per_code. The commonly-cited 16.368 MHz GPS L1 rate is ~3× slower than the nearby 16.384 MHz because of a radix-31 inner FFT. See Sampling Frequency and FFT Performance for the rule of thumb and a table of measured timings.

Algorithm Overview

This library implements the FM-DBZP algorithm from Heckler & Garrison (2009), which performs simultaneous correlation across all code phases and Doppler bins using a two-dimensional FFT structure.

The signal segment of length N = num_coherently_integrated_code_periods × samples_per_code is divided into num_blocks equal sub-blocks. A row-wise inverse FFT performs circular correlation against each PRN sub-block, forming a 2D coherent integration surface (num_doppler_bins × samples_per_code). A column-wise FFT then resolves the Doppler dimension, producing a power surface that is peak-searched for acquisition.

The implementation never materialises these surfaces: it processes one num_doppler_bins × block_size column-block tile at a time and reduces the power cells on the fly, so memory stays tile-sized (cache-resident) no matter how large the search grid is. The full power surface only exists if you request it with store_power_bins = true.

Features

  • FM-DBZP algorithm: Joint code-phase and Doppler search via 2-D FFT structure
  • Pre-computed plans: Reuse FFT plans and buffers across acquisitions with plan_acquire
  • Streaming, tile-based reduction: The search surface is processed one cache-resident column block at a time — plan memory stays in the MiB range even for wide searches, and the full power surface is only materialised on request (store_power_bins = true)
  • Multi-threaded: PRNs processed in parallel — start Julia with -t N to use N threads; extra threads cost only tile-sized scratch
  • Non-coherent integration: Accumulate power across multiple signal segments to improve sensitivity at low CN0
  • Data bit handling: Bit-edge search and sign-combination search for coherent integration spanning multiple data bits
  • Sub-sample interpolation: Parabolic interpolation for Doppler and code-phase refinement below grid resolution
  • Plotting support: Visualise acquisition results with Plots.jl

Contents