filters.weighting_compliance

IEC 61672-1:2013 frequency-weighting class verification.

A/C/Z frequency-weighting acceptance limits transcribed from BS EN 61672-1:2013, Table 3 (standard page 22): the design-goal responses and the class 1 and class 2 upper/lower limits at the 34 nominal frequencies from 10 Hz to 20 kHz. A lower limit of -inf means only the upper limit applies (subclause 5.5.6 checks measured deviations at the nominal frequencies).

IEC 61672-1:2013 defines only classes 1 and 2. Type 0, the tightest of the four instrument types, lives in the superseded IEC 651:1979 Table V, held here as the identical British adoption BS 5969:1981. Its four masks differ numerically from the 2013 edition (e.g. Type 0 is +2/-3 dB at both 16 and 20 kHz where class 1 is +2.5/-16 and +3/-inf), so the two editions are kept as separate mask tables selected by the edition argument ("2013" default -> classes 1/2; "1979" -> Types 0/1/2/3, offered as classes 0-3).

The historical B weighting is verified against ANSI S1.4-1983: design goals from the B column of Table IV (whose A and C columns equal IEC 61672-1:2013 Table 3 digit for digit) and tolerance limits from Table V, whose instrument Types 1 and 2 fill the class 1 / class 2 verdict slots. The ANSI Type 0 column is a different mask from the IEC 651 one - two-sided and stricter at 10/12.5/16 Hz where IEC 651 is upper-only - so the two are carried as the two editions they are and never merged. The AU weighting is verified against IEC 61012:1990: design goals are the sum of the nominal A response and the Table 1 nominal U response (with the subclause 2.2 explicit AU values at 25/31.5/40 kHz), checked against the Table 1 tolerances for the filter as a separate unit, the tighter of the two tolerance readings the standard offers. IEC 61012 publishes a single tolerance set, so both verdict slots carry the same margin for AU.

One subject: the weighting network a sound level meter applies to the whole signal, whose acceptance limits qualify the deviation of its measured relative response from a design goal at the nominal frequencies. The band-filter class limits of IEC 61260-1, which qualify a relative attenuation against a mask around each mid-band frequency, live in phonometry.filters.compliance.

Auto-generated from the source docstrings by scripts/generate_api_docs.py (make api-docs). Do not edit by hand.

verify_weighting_class(
    wf: WeightingFilter,
    *,
    sweep_points: int = 4096,
    edition: str = '2013',
) -> dict[str, Any]

Verify a frequency-weighting filter against its standard’s tolerances.

A/C/Z are checked against IEC 61672-1:2013 Table 3 (classes 1 and 2). The historical B weighting is checked against ANSI S1.4-1983: Table IV design goals with the Table V tolerance limits, whose instrument Types 1 and 2 fill the class 1 / class 2 verdict slots (an overall_class of 1 then reads “ANSI S1.4-1983 Type 1”). AU is checked against IEC 61012:1990: design goals are nominal A + nominal U (Table 1, plus the subclause 2.2 explicit AU values at 25/31.5/40 kHz) with the single Table 1 tolerance set for the filter as a separate unit, so both class slots carry the same margin and overall_class is 1 (complies) or None. G is not supported here (ISO 7196 defines one +/-1 dB instrumentation tolerance, no class structure; the CI conformance report pins it), nor is D (the tolerance tables of the withdrawn IEC 537 did not survive it; the conformance report pins the D response against its published transfer function and tabulated curve).

edition="1979" swaps in the tolerance table of the superseded IEC 651:1979 instead, whose Table V publishes the laboratory-grade Type 0 mask that IEC 61672-1 has no equivalent for, and three further types. Class N is then the standard’s instrument Type N, so an overall_class of 0 reads “IEC 651:1979 Type 0”. That edition covers A, B and C (the weightings of subclause 3.2, whose Table IV design goals equal the ones used above digit for digit): its Table V footnote makes one mask govern every weighting characteristic, so B does not borrow the ANSI limits there. It is a genuinely different mask, not a rename - Type 0 is +2/-3 dB at 16 kHz and at 20 kHz, where class 1 is +2.5/-16 and +3/-inf, so an error class 1 cannot see is visible under Type 0.

The filter’s relative response (normalized to its 1 kHz gain) is evaluated at the exact base-10 frequency behind each nominal label below the Nyquist frequency (IEC 61672-1 Table 3 NOTE: the design goals are computed at $f=1000\cdot 10^{0.1(n-30)}$, e.g. 15 848.9 Hz for “16 kHz”; IEC 61672-3:2013 subclause 13.3 tests the deviation at the same exact frequencies, and IEC 61012 Table 1 lists the same exact frequencies). The deviation from the design-goal weighting is checked against the two acceptance masks.

A dense logarithmic sweep between the checked frequencies additionally enforces IEC 61672-1 subclause 5.5.7: at any frequency between two adjacent nominal frequencies, the deviation of the response from the analytic design goal (Annex E for A/C/Z, the ANSI S1.4-1983 Appendix C formulas for B, the A response cascaded with the IEC 61012 Table 2 poles for AU) must stay within the larger of the two adjacent limits. Without it a resonance or notch between the nominal frequencies would go unnoticed (for B, for AU and under the 1979 edition, whose tables are tabulated at the nominal frequencies and nowhere else, the sweep is applied as the analogous engineering check). Both the per-frequency verdicts and the sweep must pass for overall_class. The sweep samples sweep_points grid frequencies; a violation narrower than the grid spacing could in principle fall between samples, so raise sweep_points for higher-Q suspects (the verdict attests the sampled grid, not a continuous proof).

The response is taken over the whole path a signal travels through filter, which is one cascade of second-order sections at the input rate for every curve and in both stateful and single-shot use. It used to be more than that: the sections were reached through an interpolation and a decimation stage whose anti-alias filter had its transition band on the input Nyquist frequency and dominated the response above roughly 0.9 * fs / 2, so a verdict read from the sections alone attested a filter the user never ran. That is why the verdict is measured through _runtime_frequency_response rather than through sosfreqz here, and it stays that way so the next stage added to the path cannot go unmodelled. The Z weighting is a flat bypass and always complies.

When rows that carry a finite lower acceptance limit fall at or above the Nyquist frequency (e.g. the 8-16 kHz class 1 rows of a 16 kHz sampled system, or the 25-40 kHz AU rows of a 48 kHz one), they cannot be checked and range_limited is True: the returned class then attests conformance over the checked frequencies only, not conformance over the standard’s full frequency range.

Parameters

Returns: Dict with overall_class (the strictest class of the edition that every checked frequency and the sweep meet, or None), range_limited (see above), bands: a list of {"freq", "class", "deviation_db", "margin_class<c>_db"} for each class c of the edition, where freq is the nominal label and a positive margin means the limits are met with that much room, and between_nominals: {"worst_freq", "margin_class<c>_db"} for the sweep.

Raises

weighting_class_limits(
    weighting_class: int,
    *,
    edition: str = '2013',
) -> tuple[np.ndarray, np.ndarray, np.ndarray]

Acceptance limits of one performance class of a weighting standard.

The limits apply to every weighting the edition defines; they qualify the deviation of the measured relative response from the design goal at each nominal frequency, not the response itself. Under edition="2013" they come from IEC 61672-1:2013 Table 3 and govern A, C and Z (the B and AU masks that verify_weighting_class uses come from ANSI S1.4-1983 Table V and IEC 61012:1990 Table 1 instead and are not returned here). Under edition="1979" they come from IEC 651:1979 Table V, whose footnote makes one mask govern every weighting characteristic, B included.

Parameters

Returns: Tuple (frequencies, lower, upper) of the 34 nominal frequencies (Hz) and the lower/upper deviation limits in dB. A lower limit of -inf means only the upper limit applies.

Raises