Thread: The GTO filter - released!

Folks, the 5.3C Firmware is ALIVE!


The short version is that it does two things:

• Minor bug fixes and optimisations
• Adds Gibbs Transient Optimised Digital (GTO) Filter (This replaces the Minimum Phase Filter.)

If you're wondering what the GTO actually is, well...

iFi audio - The GTO filter
Part 1/4 - The TL;DR


What iFi GTO is?


Based on AMR’s ‘Organic’ mode, the further refined iFi GTOTM Digital Filter resolves a number of audible problems associated with both classic digital filters AND non-oversampling systems, sounds on par (or better in optimal systems) to the latter and its transient optimised performance is similar.


How iFi GTO does it?


Our GTOTM filter tuned to fit inside the ‘ringing envelope’ of our own hearing makes ultrasonic ringing objectively measurable yet not perceived. It essentially ‘masks’ this time-domain distortion to make it inaudible.


Why do I want iFi GTO?


Because it provides natural sound free from audible time domain distortion!


Stay tuned folks! We have a lot of filter related goodies to come!

iFi audio - The GTO filter
Part 2/4 - Introducing the iFi GTO™ Digital Filter



ALL DIGITAL FILTERS FOR AUDIO ARE WRONG.
ALL OF THEM, INCLUDING THE 'NO FILTER' OPTION.
THIS IS WHY WE NEED YET ANOTHER FILTER!

All digital filters (including no filter) differ in how they are wrong and how this influences objective measured performance as well as subjective listening performance with music and indeed, specific types of music. All digital filters add specific distortion signatures in either time vs. amplitude domain or frequency vs. amplitude domain. These distortions become all the more relevant the lower the sample rate. So, the most abundant digital music source -CD quality - is most impacted with greater possible audible consequences than High-Res content.

Wherever there is a difference, there is also a preference. Subjective listening preference may be informed by a range of factors including a learned or acquired response to recorded sound (e.g. what sounds ‘right’ or ‘hifi’ is not what sounds natural in comparison to a live performance), including direct referencing acoustic music performances.



However, with sufficient data from extensive listening tests and some inductive thinking, one should be able to propose and implement a digital filter that offers substantial improvements in removing ultrasonic noise over the ‘no filter’ (non-oversampling) case while avoiding as much as possible erring too far in the other direction with excessive and audible ringing.

So here it is - the ever so musical iFi GTO™Digital Filter in the Pro iDSD which is the first ever seen in any DAC. In due course, technological hurdles permitting, we will try to implement it as a firmware upgrade for just about all[1]iFi audio digital products.

What is the iFi GTO™ Digital Filter?

The Gibbs Transient Optimised filter (GTO) is named after the ‘Gibbs phenomenon (2)’ in mathematics.

Wikipedia referred to the Gibbs phenomenon as “the peculiar manner in which the Fourier series of a piecewise continuously differentiable periodic function behaves at a jump discontinuity. The nth partial sum of the Fourier series has large oscillations near the jump, which might increase the maximum of the partial sum above that of the function itself. The overshoot does not die out as n increases, but approaches a finite limit.”

Most crucially, this is one cause of ringing artefacts’ in signal processing which the GTO addresses.

Way back in May 2011, the parent company of iFi audio, AMR, pioneered an earlier version of this filter in the DP-777 digital processor where it was available as an ‘Organic’ filter. Since 2011, more time has been invested into producing a filter that offered both better compatibility and technical performance than non-oversampling, while delivering a transient optimised performance that differs as little from non-oversampling as possible, delivering the new GTO™ filter.



Non-oversampling Transient response vs Organic- Digital Filter AMR DP-777

No doubt there will be extended debate if our GTO™digital filter offers the right trade-off, compared to others. To us the two key qualities we sought was to shape of the unavoidable transient or time domain distortion so that is free of any ‘pre-ringing’ and that completes its impulse response within a fraction of the Haas (precedence effect) window; to remain in effect, inaudible to the human ear.

What we really refer to when we are talking about ringing in digital filters is actually a form of ‘Echo’ or ‘Reverb’ where,in addition to the actual transient time-shifted lower amplitude, copies of the impulse are generated using delay lines(see also the transients and digital filters section later on).

The human hearing itself is subject to an inherent transient post (impulse) ringing that completely decays within around 0.7mS[1] (see also the transients and the human hearing section later on).

The GTO filter’s transient post-ringing decays completely within 0.72mS for a 44.kHz source,ensuring that the unavoidable blurring of the transient response cannot be heard, but is integrated by the human hearing into the original transient.

This is in stark contrast to some alternative filter concepts. For example, the ‘Transient Aligned’filter seeks a maximum number of taps, leading to an impulse response that falls well outside the Haas window. ie. its ‘ringing’ is very audible, in part because there is a pre-ringing (or pre-echo) present and in part through the sheer length of the delay line used.

For example, the 16k tap Transient Aligned filter in the Pro iDSD has an impulse response with equal pre- and post-ringing trail of around 186mS @ 44.1kHz sample rates, or a total 386mS worth of ringing. This is certainly sufficient time delay to be perceived as reverb. Using an even larger number of taps lengthens this impulse response even more.




Transient Aligned Digital Filter Transient response vs GTO™ Digital Filter iFi iDSD Pro

It may be of course, that some will prefer the sound of a very long filter, with large amounts of ringing/reverb/echo as the result is often perceived as extra added spaciousness, however, to anyone seeking to be close to the original musical performance such additives are usually unwanted.

In the end, with the iFi GTO filter, by keeping the filter short and without pre-ringing, the filter response is inaudible because it is masked by the limits of the human hearing system. At the same time this filter still permits significant attenuation of unwanted ultrasonic images, compared to non-oversampling and also other attempts at “low tap number digital filter”.

Analogy: if a 20million mega pixel camera was used to take a picture of a straight line, the naked eye would see only a straight line. As the resolution is ‘beyond’ that of the human eye, any ultra-fine imperfections are not ‘seen’. This is the same as with the GTO filter withhuman hearing.

If the GTO™digital filter is so ‘perfect’, why include the other filters with the Pro iDSD? As remarked before, individual listeners may have different listening preferences and rather than imposing one option, even if we feel this option is not the best, we prefer to leave the choice down to the individual.

1) The original iDAC micro cannot receive this upgrade
2) https://en.wikipedia.org/wiki/Gibbs_phenomenon
3) “Response of the human tympanic membrane to transient acoustic and mechanical stimuli: Preliminary results” Payam Razavi, Michael E. Ravicz et al - Hear Res. 2016 Oct; 340: 15–24.

Originally Posted by Elberoth

What is the difference between the GTO filter and other popular minimum-phase filters, that eliminate the pre-ringing by the reconstruction filter ? Shorter post-ringing ?

The key differences are in our goals and starting points and thus the results.


Most of the minimum filters evolved from a desire to minimise ringing as much as possible. This comes at the cost of rejection of ultrasonic images of the wanted signal, being weak to very week in most cases. In other words, these filters tend to be "too minimal".


Our GTO filter evolved from the opposite point, namely our continuous use of filterless DACs implemented from the very first AMR/iFi product - the CD-77.


So our point was literally the opposite, we wanted to add material rejection of ultrasonic noise to non-oversampling DACs without the problems caused by analogue filters steep enough, while retaining the sound characteristics of a non-oversampling system.


Our filter is designed to offer the maximum out of band signal suppression compatible with making the filter inaudible to the best of our research and in-house listening tests.


So our filter rejects unwanted, higher order ultrasonic images more than the other "minimum" options know to us, while remaining inaudible.

iFi audio - The GTO filter
Part 3/4 - Introducing the iFi GTO™ Digital Filter

How taps relates to what is heard

So far, we have identified that we prefer the GTO filter because it has few taps.
Because:

More taps = more reverberation.
Few taps = minimal reverberation

Reverberation[1] is artificial. Sound engineers add reverb to make recordings more spacious, artificially so. Digital filters introduce reverb by the nature of their operation. In fact, a digital reverb unit operates precisely like a digital filter in principle – as depicted in this diagram.

Within digital filters are Digital Delay Lines which is defined by Wikipedia:

"A digital delay line is a discrete element in digital filter theory, which allows a signal to be delayed by a number of samples.
Delays of N samples is notated as {z} ^{-N} motivated by the role the z-transform plays in describing digital filter structures.
Digital delay lines are widely used building blocks in methods to simulate room acoustics, musical instruments and digital audio effects."

To our ears, the GTO filter simply sounds ‘right’ without any hint of artefacts or exceptional detail that feels ‘processed’, by avoiding large number of tap’s that add excessive reverb.



Transients and the human hearing

The human ear is a marvelous system with an incredible dynamic range (~135dB in middle frequencies) huge bandwidth (almost 1:1000) and a transient resolving ability that exceeds the upper limit of hearing steady state tones.Yet it is also subject to limiting factors which result in, so to speak, “blind spots” in its behavior that do not exist in purely mechanical systems (e.g. microphone). These “blind spots” can mask some behavior which objectively is distortion to be inaudible. For example, harmonic distortion masking has been well documented since the at least the 1950s if not earlier and it is reasonable to consider that ‘ringing’ on transients is also masked to a certain degree.

If we wish to produce audio gear that is capable of operating in a way that subjectively sounds undistorted to the human hearing (the most logical preference), we must understand its limitations and capabilities. Here, we focus on the time-domain capabilities.

It has been shown that the human hearing’s time domain resolution for the initial transient may be as small as 5μs. Some debate remains as to the exact limits, though work done by Dr Peter Lennox of Derby University suggests a median between 13…18μs, or a location accuracy of less than 2 degrees.



Additionally, the transient response of the human hearing includes 500...700uS ringing caused by the ear’s mechanical system
(Tympanic Membrane, Malleus/Incus/Stapes).



This ringing occurs after a transient event, there is no pre-ringing. The ringing in the ear’s system will mask any similar external ringing, which will instead be integrated into the transient, so it is inaudible.

Any pre-ringing is not masked by the human hearing, nor is any ringing that continues substantially beyond 500...700uS.