dc.title: "OA-OBBQ Test Case Set: Overlap-Add Block-Based One-Bit Quantization for Linear-Phase FIR Filters" dc.creator:

  • "Mayer, Florian" dc.contributor:
  • "Vogel, Christian" dc.subject:
  • "one-bit quantization"
  • "block-based optimization"
  • "noise shaping"
  • "linear-phase FIR filter"
  • "minimum-phase FIR filter"
  • "overlap-add"
  • "mixed-integer quadratic programming" dc.description: > Parameter set defining the synthetic test cases used to evaluate the overlap-add extension of the block-based one-bit quantization framework (OA-OBBQ) for linear-phase FIR shaping filters. Each case fully specifies one test scenario: the multitone input signal, the noise-shaping filter, the reconstruction filter and the block size. The cases are grouped so that each group varies a single design axis while all remaining parameters are held fixed. dc.publisher: "FH JOANNEUM - University of Applied Sciences, Graz, Austria" dc.date: "2026-08-24" dc.type: "Dataset" dc.format: "text/x-python" dc.identifier: "testCases.py" dc.source: > Mayer, F. and Vogel, C., "Block-Based Optimization for Frequency-Selective One-Bit Quantization", IEEE International Symposium on Circuits and Systems (ISCAS), 2025. dc.language: "en" dc.relation:
  • "Mayer, F. and Vogel, C., ISCAS 2024 - optimization-based one-bit quantization (OBAQ)"
  • "Mayer, F. and Vogel, C., ISCAS 2025 - block-based one-bit quantization (OBBQ)" dc.coverage: "Synthetic multitone signals, normalized frequency 0 to pi, N = 2048 and 4096 samples" dc.rights: > Research output, shared under CC BY 4.0. Funded by the Austrian Science Fund (FWF) [10.55776/DFH 5] within the DENISE project and the province of Styria.

OA-OBBQ Test Case Set

Contents

testCases.py holds a single Python list, lCases. Each entry is a dictionary that fully specifies one test scenario. Running the generator over this list produces one signal batch per entry; every batch contains sBatchSize independent realizations of the same scenario, differing only in the random draw of tone phases and frequency bins.

A scenario has four parts:

  • the input signal — a multitone signal with energy confined to the band(s) mWX, normalized to the drive level sBound
  • the shaping filter — the FIR filter whose matrix defines the cost function of the quantizer, specified by band(s) mWD, length sL and the Kaiser design parameters
  • the reconstruction filter — the FIR filter used to evaluate the result, specified by band(s) mR; independent of the shaping filter
  • the block size sM — the number of bits optimized jointly

Variables

Variable Type Meaning
strSig str Signal type. "real" denotes a real-valued multitone signal.
mWX ndarray, n×2 Signal band(s). One row per band, [omega_lo, omega_hi] in radians per sample, range 0 to pi.
vAmp ndarray or None Amplitude per tone. None draws them randomly.
vPhase ndarray or None Phase per tone in radians. None draws uniformly from 0 to 2 pi.
bUseCos bool Excitation function. True = cosine, False = sine.
sKRatio int Spacing of the occupied frequency bins within the band. 1 occupies every bin.
strKMode str Distribution of the bins. "lin" spaces them linearly across the band.
bReplace bool Whether bins are drawn with replacement.
sBatchSize int Number of independent realizations generated per case.
sN int Signal length in samples. Must be an integer multiple of sM.
sM int Block size, i.e. the number of bits optimized jointly per block. The number of blocks is sN / sM.
sL int Length of the shaping FIR filter in taps.
sBeta float Kaiser window beta of the shaping filter. 0.0 lets the design derive it from sAsb.
mWD ndarray, n×2 Passband(s) of the shaping filter, same format as mWX.
mR ndarray, n×2 Passband(s) of the reconstruction filter, same format as mWX. Independent of mWD.
sBound float Drive level. The signal is normalized to the range [-sBound, +sBound].
kaiser dict Filter design parameters, see below.

The kaiser entry holds three fields:

Field Type Meaning
sApb float Maximum passband ripple in dB.
sAsb float Minimum stopband attenuation in dB.
sDeltaW float Transition width in radians per sample.

All frequencies are normalized angular frequencies in radians per sample; pi corresponds to half the sampling rate.

Groups

The list is ordered into eight groups. Within each group exactly one design axis is varied and all other parameters stay at the baseline value.

Group Varied axis Values
baseline — sM = 32, sL = 79, band 0 to pi/10, 90 dB stopband
A block size sM = 8, 16, 32, 64 at sL = 79
B filter length sL = 127, 213, 341 at sM = 32, transition width scaled accordingly
C band position and width cutoff pi/4, cutoff pi/32, bandpass pi/8 to pi/4
D stopband attenuation 40 dB and 120 dB against the 90 dB baseline
E signal band vs. shaping band signal narrower (pi/40), wider (pi/6) and offset (pi/12 to pi/10) at a fixed shaping band of pi/10
F multiband shaping two disjoint shaping bands, 0 to pi/16 and pi/4 to pi/3
G reconstruction mismatch mR wider (pi/8) and narrower (pi/14) than mWD
H transition width, drive, phase transition pi/6, sBound = 0.5, sine excitation

Note that mWD and mR are independent parameters. They coincide in most cases, but group G deliberately sets them apart: the quantizer minimizes the error weighted by the shaping filter, while the evaluation weights it by the reconstruction filter, and the two measures coincide only when the filters are identical.

License

Shared under CC BY 4.0.