[USAudio]

There are many products hitting the market these days by companies such as Ayre, Meridian, etc., that incorporate "minimum phase" digital filters that prevent the introduction of pre-ringing and lessen post-ringing.

Can Dan or anyone in the know here detail if any of the Lavry products (current or future) deal with this issue and, at a high-level, how?

For reference, here's a nice short informative paper from Ayre describing the issue and how they decided to deal with it:
http://www.ayre.com/PDF/Ayre_MP_White_Paper.pdf

Thanks!

 

[wavoman]

Great question! More details on this issue would be great.

Can't you also totally eliminate the pre-ringing problem with a filter that chooses to e.g roll off the highs on purpose -- i.e., if you are willing to mess with the linearity of the mapping of frequencies in the recovered analog signal then you can have near-perfect transient recovery? Isn't this what Wadia did back in the day that made them famous?

People my age can't hear the highs anyway, but we can hear a muddy transient.

Maybe Dan or any of the real engineers here can separate the fact from the hype in this pre-ringing issue for us ... we're all ears (to make a pun).

 

[Dan Lavry]

Quote:

Originally Posted by USAudio /img/forum/go_quote.gif There are many products hitting the market these days by companies such as Ayre, Meridian, etc., that incorporate "minimum phase" digital filters that prevent the introduction of pre-ringing and lessen post-ringing. Can Dan or anyone in the know here detail if any of the Lavry products (current or future) deal with this issue and, at a high-level, how? For reference, here's a nice short informative paper from Ayre describing the issue and how they decided to deal with it: http://www.ayre.com/PDF/Ayre_MP_White_Paper.pdf Thanks!

The subject of filters goes much beyond the technical expertize of most people in this forum. Understanding filters calls for good math background and much study of EE material.

As a rule, most filters out there are minimum phase. I am for linear phase as well, because linear phase means an EQUAL DELAY for all the frequencies, thus it means RETAINING the original waveform intact.

Oner can make a real poor filter where pre ringing and post ringing is audible. But one can also make a good FIR that is not auible. One does not hear a filter, one hears the INTERACTION of a filter and the musical material (the data fed to the filter). It is the OUTCOME of that interaction that needs to be looked at. For example, what is the frequency of the pre ringing? Is it audible under any input signal?

My filters are all minimum phase and as I stated, I am for linear phase. A non linear phase alters the waveform, because it delays the fundumental and various harmonics by different amounts of time. Say you take a 1KHz square wave (with harmonic content to say 22KHz). You can look at it as a sum of a 1KHz sine wave and various harmonics (3Khz, 5KHz, 7KHz...21KHZ). For a square wave, the even harmonics are zero amplitude.

Now, if you delay the fundumental and all the harmonics by the same amount (linear phase), you end up with the SAME sqaure wave, only delayed in time. But if you delay the various components (fundumental and harmoics) by different amounts of time, the waveform is no longer the same!

Some will argue that the ear does not care. Others will argue that the ear does care. But no one can argue that the wavform is the same. It may turn out to have slower rise, higher amplitude or what not. So the analog circuits will see a different waveform, and at this point, I care! I am very much into preserving the waveform because my goal is transparancy.

Of course, how linear is linear? In the case of FIR one can make phase linearity perfect. For analog filtering, we deal with some compromises, and for a NOS, phase is a real problem. For upsamplinf DA's one can provide good phase linearity.

Regards
Dan Lavry

 

[USAudio]

Thanks Dan. I was just curious if the filters you employ in your products (such as the DA11, which I own) do not introduce pre-ringing, such as what the products I listed above claim. It's my understanding that this unnatural pre-ringing is what has given many products through the years that undesirable "digital" sound, which the Lavry products don't seem to suffer from.

Based on your statement "One can make a real poor filter where pre ringing and post ringing is audible.", I assume your filters don't.
Can you expand on that statement anymore for us neophytes?

Thanks again!

 

[Dan Lavry]

Quote:

Originally Posted by USAudio /img/forum/go_quote.gif Thanks Dan. I was just curious if the filters you employ in your products (such as the DA11, which I own) do not introduce pre-ringing, such as what the products I listed above claim. It's my understanding that this unnatural pre-ringing is what has given many products through the years that undesirable "digital" sound, which the Lavry products don't seem to suffer from. Based on your statement "One can make a real poor filter where pre ringing and post ringing is audible.", I assume your filters don't. Can you expand on that statement anymore for us neophytes? Thanks again!

The subject requires knowhow and it is not easy to put into simple words.

There are two very basic "types of filters", analog filters and digital filters. Here we are talking about digital filters, and there are two basic types of digital filters, FIR and IIR.

The IIR (infinite impulse response filters) are in fact an "emulation" of analog filters in the digital domain. Such filters are not perfect emulation, but they tend to have near similar response as their analog cousins. The process of transforming from analog to digital is normally called z transforms, and the mapping the frequencies from digital to analog introduces some differences in the frequency axis at the very low and the very high frequencies (such as near 100Hz and below and also near 22KHz for a 44.1KHz system). So one needs to pre warp the frequencies before the transformation. It gets pretty complex…

One can make a nice IIR, but it is also easy to do a poor job of it. There are a lot of things that can go wrong. For example, an IIR can have a lot of unwanted low level activity which is called limit cycles or fractals. A serious designer will comply with what is needed to make sure that the limit cycles do not happen, but some designers just do the Z transformation without any awareness of the fine details of the art of IIR design. One needs to have some serious specific expertise.

Like analog filters the use of IIR brings about the issue of phase response. Like the analog filters they emulate, IIR's come in all sorts of flavors, each one with it's own set of advantages and features. The most common types are Butterworth, Bessel, Chebychef… Each type offers some desired feature such as “best for flatness”, “best for phase”, closer to "brick wall" and so on. Each of the types can have few or many "stages" (poles and zeros), such as 2nd order, 3rd order... 5th order.... For analog filters, one can chose various circuit implementations such as multiple feedback, Sallen Key, leapfrog, and many more! Similarly, for IIR, one can choose from an array of implementations as well (we call it “topologies”).

The use of IIR is growing in digital audio, especially in AD’s mostly because they offer less delay in the signal path (lower latency). Latency may be an issue when a player recording a track wants to hear it in real time. Too much delay between the time you “hit a note” to the time you hear (through an AD then DA…) it is not a good thing. So IIR’s are one way to make for faster delay. While IIR can offer no pre ringing, that is not the reason for using it, the reason is, as I stated, lower latency (delay).

But when gear is not used in critical recording while self monitoring applications, having some delay is a non issue. You press to “play button”, and if the music is played 10msec later, one does not care. So for ideal listening, one can overcome many of the IIR shortcomings by use of FIR digital filters (finite impulse response filters). One can make a linear phase FIR, and there are many ways to make FIR’s. I have a paper about understanding of the basics of FIR’s my site at Lavry Engineering. FIR’s can be designed to excel at various tasks, and as a rule, the length of the FIR (number of coefficients) has much to do with the filter capabilities.

There are different ways to design an FIR. One can design an FIR by imposing a “window” on a sync function (which is an ideal impulse response), and there are many types of windows to chose from, each with it’s own advantages and strength. A good window design can yield a great FIR, but such designs are not ideal from a required processing power standpoint. To solve that issue, many designers and filter design software offers optimizing techniques help yield acceptable results that call for less compute power. In my opinion, many of the poor filters are a result of cutting corners to save some compute power.

So at the end of the day, filter design is a very complex and involved subject, and the designer should know both their end goal, and all that it talks to get there. I barely touch the subject and the post is already so long… Sorry to say, a single buzz word, be it “minimum phase”, “linear phase”, “pre ringing” and so on is not much more then a buzz, that stuff is good for marketing. The “pre ringing” and “gradual filter” is not all that new either, and the concepts have been around for over a dozen years. A “new slew of converters with minimum phase” is nothing but marketing. I am pretty sure that all the DA filters out there are minimum phase…

If you want a more pointed response, I will try to accommodate, but the subject of pre-ringing is rather complicated. It would be good if you can be equipped with some knowledge of FIR's.

Regards
Dan Lavry

 

[USAudio]

I guess I was curious as to how specifically the DA11 does with pre-ringing but at the same time certainly don't expect you to give up any trade secrets or implementation details.

 

[Dan Lavry]

Quote:

Originally Posted by USAudio /img/forum/go_quote.gif I guess I was curious as to how specifically the DA11 does with pre-ringing but at the same time certainly don't expect you to give up any trade secrets or implementation details.

In order to understand the issues of pre ringing and post ringing, one has to be equipped with enough knowhow in the area of digital filters (specifically FIR), as well as understanding of is audible by the ear.

I already mentioned earlier that one does not hear a signal that is shaped like the filter coefficients itself. When you look at the coefficient curve, which looks very similar to like a sinc function, you see the shape of "ringing" on both sides of the main lobe. But that is NOT what you hear. Why? Because the output of an FIR at each sample time is the SUM of the PRODUCTS of the coefficients and the input signal. Your signal (data) is NEVER a single unity impulse preceded and followed by zeros. That is not a real world signal, such signal represents infinite bandwidth.

So what is a real signal look like? Well, just about anything, as long as it fits within the audible bandwidth (less then Nyquist). A sequence such as ...000010000... is "out", and that is the only way to hear the filter "coefficients shape" of the filter. So how about sequence such as ....0000001111111....? That is also not real. A sudden step is also infinite bandwidth. So you can see that the plot of the coefficients is not representing the output waveform. Figuring the output for real signals is much more complex, and at the end you have to make sure that you end up with non audible result from a pre-ringing and post-ringing standpoint.

One has to examine the signal of the FIR output for the most extreme real world signals (band limited to Nyquist), and make sure it is not audible. For example, if there is some ringing, and it happens at say 20KHz, will you hear it? It depends on the nature of the signal - the cycle frequency, the duration of the ringing and the "envelop" of the signal plays a key role here.

I do not want to say much more, what I do in my DA is proprietary. I hope that what I said is enough to point out that the subject is not as simple as it first appears. At the minimum, one should realize that a quick glance at a coefficient curve is far from sufficient view of the issue.

Regards
Dan Lavry

 

[USAudio]

Quote:

Originally Posted by Dan Lavry /img/forum/go_quote.gif I do not want to say much more, what I do in my DA is proprietary. I hope that what I said is enough to point out that the subject is not as simple as it first appears. At the minimum, one should realize that a quick glance at a coefficient curve is far from sufficient view of the issue.

I figured we were getting into the proprietary territory but you've answered my question in that clearly you've given pre/post-ringing a lot of thought in the design of your DACs.
Thanks for taking the time to share your thoughts and expertise Dan.

When is the "mainstream" audio press going to give the DA11 or DA10 as much coverage as they give the DAC1? Sure seems lopsided when almost every owner review I've read has expressed a preference for the sound quality of the Lavry DAC.