Same Filter, Different Results
Brand Variations in IIR EQ for Loudspeakers & Audio Systems
Infiinite Impulse Response (IIR) filters have been used for decades to EQ loudspeakers or entire systems.
Most processors have user-accessible settings to accept IIR filters (in the form of data either manually entered on a hardware interface, or entered via compatible control software).
An ever-increasing number of processors also now accept user-generated Finite Impulse Response (FIR) filters. FIRs & IIRs each have their advantages and it’s often beneficial to use a combination of both (learn more).
Common IIR filter types (available for use in most amplifiers & processors) include:
- Parametric
- Notch
- Band-pass
- All-pass
- Low & high shelf
- Low-pass & high-pass (Bessel, Butterworth, Elliptic & Linkwitz/Riley)
- Low-pass & high-pass with variable Q
Unfortunately, identical IIR performance across different processor brands using the same IIR filter cannot be guaranteed, as some filters produce different results when used across processor brands.
Common causes for these differences are variations in the way each brand’s software interprets some filter parameters, like Q and bandwidth (BW), and in the conversion between Q & BW. The differences typically appear in parametric (peaking) filters, shelf filters and all-pass filters. (For example, a parametric filter on a Linea Research processor doesn’t match a parametric filter on a Powersoft amplifier when using the same parameters.) There can also be differences between Bessel filters where brands have different interpretations of rolloff frequency; ‘natural’ or ‘-3 dB’. Different products also have different IIR filter parameter limits.
The plots below show various implementations of a 10 kHz 10 dB parametric filter. The upper row is for a sampling rate of 48 kHz and the lower for 96 kHz. There is no simple translation between the BW and Q in each implementation, so to help with the comparison, all have been manually & approximately matched by adjusting the BW or Q to give ~ +5dB at 5 kHz. Looking at both the magnitude and phase, it’s clear that the implementations are different, particularly near the Nyquist frequency of 24kHz for the 48 kHz sampling rate. The 96 kHz implementations are more similar because Nyquist has been pushed out to 48 kHz. However even when overlaid there are subtle differences.
In the plots, the implementations in the right two columns (labelled Linea Research & Powersoft) each have unique equations to create 2nd order analog filters, which are then mapped to 2nd order digital designs using the Bilinear Transform with frequency warping. The plot on the left uses a method proposed by Sophocles Orfanidis which attempts to keep the digital filter shape close to the analog design near the Nyquist frequency. (This method is very similar to the implementation used by Lake Processing). For more details see the 1996 AES Paper “Digital Parametric Equalizer Design with Prescribed Nyquist-Frequency Gain.”
These differences don’t typically affect audio professionals who only ever work with systems using processors from a single brand (or OEM manufacturer).
However, sometimes it’s necessary (or desirable) to use different brands of processor side-by-side within an installation, or it’s necessary to duplicate IIR settings from one brand in another (for example, when equipment suppliers are changed or particular audio system components are upgraded).
In addition, when EQ’ing a new system (in the same or a new location), independent installers, system/monitor engineers and technicians often start with baseline IIR settings developed previously, and adjust from there. But when new processor brands are involved, this baseline can shift. Duplicating IIR settings from one brand of processor in another won’t guarantee the same results, which can ultimately impact performance.
Until recently, the only solution to this problem has been an (often lengthy) trial-&-error process to manually adjust existing IIR filters developed for ‘Processor Brand A’ to compensate for the differences in how ‘Processor Brand B’ will interpret those same filters.
If simultaneous access to both Brand A & B is available, then using an audio measurement tool like Rational Acoustic’s Smaart® to measure & compare responses is helpful when making these adjustments. However, where access to both processors isn’t available, its near-impossible to adjust an existing IIR response so that it performs the same on the second processor as it did on the first.
However, Eclipse Audio has developed solutions to this problem, including built-in IIR brand functionality and automatic IIR filter set conversion. These features are available in Eclipse Audio’s ‘FIR Designer’ range of DSP design software tools. (Learn more and read what users have to say).
