Apodising (& Digital) Filters used in CDPs

In the other thread, Bombaywalla said:

The pre-ringing - Not weird at all if you understand DSP or maybe brush up on it. Why don't we start a new thread on this topic (apodising & digital filters) since this is Bhagwan69's ad to sell off his Meridian CDP??

I agree, the reason I called it weird was that it is weird in the sense that it is not 'natural' and probably no other distortion adds distortion BEFORE a signal (that causes the distortion ! ) is played....

Ampnut....your stage now....

Would love to hear about "apodising" and the other stuff.....

Manek

It seems that ALL CD players since the beginning of time, use digital filters that cause 'Pre Ringing' ( Its truly weird... like a time machine ! )

Amp_Nut, actually it's not so weird when you understand what's going on: The pre-ringing & post-ringing that these folks are talking about is an artifact of Gibbs phenomenon that is present in all band-limited linear-phase filters. All band-limited filters produce a Sinc function in the time-domain as per Shannon's theory. It's just the physics/characteristic of such filters.

The digital filters that have been traditionally used in CDPs are what are called 'linear phase filters'. These filters are also called Finite Impulse Response or FIR filters. One of the stand-out characteristics of FIR/linear-phase filters is that ALL frequencies in the pass-band are delayed by an equal amount thru the filter. From a music view-point this is really fantastic because the digital filter does not add any phase distortion. And, since phase & time are duals of each other, no added phase distortion means that the timing of the music signal is un-disturbed.

Now, from a EE perspective, a filter's characteristic is usually studied by what is called an impulse response. A signal of finite amplitude & a very, very small time duration - an impulse - is input into the filter & the way this impulse is shaped by the filter tells us what the filter is doing to the input fed impulse.
How small is the time duration? Generally 100 to 1000 times smaller than the filter time-constant. For example, if an audio filter is the object of study, then, a impulse that is 100X larger than 20KHz would serve as an appropriate impulse.

Traditionally, the linear-phase digital filters used in CDPs have been very sharp roll-off filters - the pass-band is 20Hz->20KHz & then a huge amount of attenution between 20KHz & 22.05KHz, which is the half-band/Fs/2 of the 44.1KHz sampling frequency. For all practical purposes, this digital linear-phase filter looks like a brick-wall filter & it defines a (very) band-limited filter. If you take the fast fourier transform (or FFT. This FFT translates from frequency domain to time domain) of such a (very) band-limited filter, you will pre-ringing & post ringing. Equal amounts of pre & post ringing.
What this basically says is that every time there is an abrupt change in frequencies, the filter output will show pre & post ringing.

If the near-vertical edges of the filter at 20Hz & 20KHz are made slopey, the amount of pre & post ringing will reduce. However, since the filter transition band (from pass-band to stop-band) is gentle, the frequencies beyond the half-band (or Fs/2) are not attenuated enough & still have sufficient energy & mix back into the 20-20KHz audio band. This is known as "aliasing". Such filters are termed as "minimal phase soft-knee filters".
The very fact that they have "minimal phase" in their name suggests that there is some phase distortion in that the delay of frequencies in the audio band are not equal thru such a filter. The trade-off made in such filters is that they introduce some phase distortion to reduce the pre-ringing. Further, if the excursion from pass-band to stop band is gentle (soft-knee filter) then the amount of post-ringing is also reduced.
So, one has to basically pick one's poison:
* linear-phase filters - they do not disturb the music signal's timing but introduce pre & post ringing due to their band-limited nature. These filters are said to contribute to digital hash that gives CD audio a bad rap of its digital characteristic.
* minimal phase, soft-knee filters - they introduce phase distortion which corrupts the timing of the original music signal & they introduce aliasing but they reduce/eliminate (depending on implementation) the digital hash by eliminating pre-ringing & reducing post-ringing. The music might sound more analog-ish but doesn't have the verve that is expected of the instruments, artist.

However for ALL CD players to play a 'Preview" before the actual note.. WEIRD and some point this to one of the reasons for 'Digital" sound.

Hopefully, the above explanation helps understanding this phenomenon.

The Meridian and the Ayre CD players now build in a filter that practically eliminates this "Pre Ringing" but at the cost of a longer 'Post Ringing' its just 1 of the many Very innovative technologies in this 'Super Player'...

The Ayre & Meridian CDPs seem to use a filter known as an "apodising filter". In an apodising filter, the pass-band of the filter is not flat all the way to 20KHz (as it is in linear-phase filters); rather, the high-frequencies of the filter are rolled-off well before the half-band (or Fs/2) frequency such that an apodising filter fully attenuates by 18KHz. Obviously, if the filter has fully attenuated by 18KHz, the roll-off of high frequencies must have begun well before 18KHz. By fully attenuating by 18KHz they are preventing the phenomenon of aliasing which is characteristic to minimal phase soft knee digital filters BUT you pay the price of a rolled-off top-end.

So, you can see that each filter has its pluses & minuses & that there is no do-it-all filter.
Thus, what you will find is that expensive CDPs like the Meridian, the Wadia 381 & maybe the Esoteric & EMM Labs using a combination of digital filters:
* minimal phase filter to eliminate the pre-ringing
* soft-knee filter to reduce the post-ringing &
* apodising filter to eliminate/reduce the effects of aliasing created by the soft-knee filter.

Most of you would have heard of Wadia's "DigiMaster" algorithm? That algorithm was supposedly developed in conjunction with Pioneer & is also called the Wadia/Pioneer Legato filter system. From a pix I saw of the impulse response, this filter has both pre & post ringing but both have been severely reduced compared to a traditional linear-phase filter. Thus, phase info of the music signal is less corrupted than a comparative minimal-phase filter & the effects of aliasing are also minimized compared to a soft-knee digital filter. Thus, one ends up with very good timing & very good imaging - the best compromise.
Here's what the Wadia/Pioneer Legato filter impulse looks like (courtesy Werner from Belgium):



Hopefully things are a bit clearer now Amp_Nut & you have shed your thoughts that CDPs have time-machines built into them. LOL!

Here is a discussion of the same by Charles Hansen of Ayre and some AA members.

http://www.audioasylum.com/cgi/t.mpl?f=digital&m=148231

Siva had mentioned that we need not put any filters..if the rest of the CDP is properly design and implemented then why does one need this filter ? what is the value that it brings in which makes High end designers put it in their CDP ?

sivat schrieb:

Amp_Nut schrieb: As a techie and engineer, I came across some Unique tech info on this CD player that I wanted to share...This CD Player's 'Apodising Filter' It seems that ALL CD players since the beginning of time, use digital filters that cause 'Pre Ringing' ( Its truly weird... like a time machine ! ) Pre Ringing creates a ringing or oscillation / vibration BEFORE a note is played, and a similar ringing After the note ends. The Post ringing can be deemed as 'natural' in that the string's sound when replayed continues even after the original stops... like an over hung sound. However for ALL CD players to play a 'Preview" before the actual note.. WEIRD and some point this to one of the reasons for 'Digital" sound. The Meridian and the Ayre CD players now build in a filter that practically eliminates this "Pre Ringing" but at the cost of a longer 'Post Ringing' its just 1 of the many Very innovative technologies in this 'Super Player'... So why use digital filter at all ? . Build the DAC right into the CD player with I2S....and if you want any further "jing-bang", just reclock with precision - to reduce the impact of jitter. (My next project !!) Infact, i do not even use any filter after the analog conversion is done...an advantage that will be strictly restricted to DIY'ers

Arj, it's my opinion that Sivat's statement of not using a filter pertains specifically to a non oversampled DAC. In such a case the DAC chip is running at 44.1KHz & the half-band (or Fs/2) freq is the DAC bandwidth automatically.


If you do not have a non oversampled DAC then, in a sampled system you simply CANNOT do this i.e. not use a band-limiting/anti-aliasing filter before that DAC.
Most of the DACs that are available probably have a band-limited/anti-alias filter built into the same I.C./package as a composite structure. In the worst case a DAC would have a sample-and-hold (S&H) circuit in front of it - that S&H is a zero-order hold circuit which is basically a filter!

Re. the I2S connection: It's my opinion that sivat is talking about an efficient connection of the transport to the DAC using the I2S connection (which calls out a bit clock, a word clock & a data line as the interface between transport & DAC). Whether you use the I2S or any other connection between transport & DAC, one needs to band limit the digital signal coming off the transport otherwise you will have aliasing galore & a horrible reproduced sound - there is a lot of digital crXX in the raw digital signal coming off the transport.
If one uses a non oversampled DAC then the band limiting is automatic & if one uses an oversampled DAC then a digital filter is necessary.

Arj schrieb:

what is the value that it brings in which makes High end designers put it in their CDP ?

And, that is the value of this digital filter to any CDP.


sivat further mentions not using any filter after the analog conversion. This time around the filter he's talking about is an analog filter (the one before the DAC is a digital filter) & you CAN totally avoid this analog filter if you so wish - you see this all the time in non oversampled DACs commercially available. You also see this in DACs using a vacuum tube output stage where the limited bandwidth of the tube acts like a slow roll-off analog filter.

bombaywalla schrieb:

Whether you use the I2S or any other connection between transport & DAC, one needs to band limit the digital signal coming off the transport otherwise you will have aliasing galore & a horrible reproduced sound - there is a lot of digital crXX in the raw digital signal coming off the transport.

Ah!

bombaywalla schrieb:

sivat further mentions not using any filter after the analog conversion. This time around the filter he's talking about is an analog filter (the one before the DAC is a digital filter) & you CAN totally avoid this analog filter if you so wish - you see this all the time in non oversampled DACs commercially available. You also see this in DACs using a vacuum tube output stage where the limited bandwidth of the tube acts like a slow roll-off analog filter.

Any reputed products that does this ?....i'm very eager to know.

sivat schrieb:

Any reputed products that does this ?....i'm very eager to know.

what is your definition of "reputed products"??

Juat FYI guys (& maybe you already know this) but my research is showing that the term "apodising" seems to come from Astronomy where apodising optical filters have been used for a long time to view dim stars in the presence of much brighter stars. In astronomy apodising filters create a non-uniform illumination profile that goes to zero (i.e. becomes black, no light) at the edges. These optical apodising filters reduce the resolution of the image but since they reduce the diffraction effects at the edges they can enhance small details (such as allow the observer to see a dim star in the presence of brighter stars).

For the EEs (Amp_Nut) - apodising filters have been in use for a long time but probably been unaware of this fact as these filters have been always called "windowing functions"!! When a FFT is done, before the results are displayed, the FFT is windowed using a Hanning/Hamming/Blackman/Kaiser/Rectangular window to remove/reduce the discontinuities at the boundaries of 1 period of the waveform. Remember that? Well, these windowing functions are actually apodising filters - they create a non-uniform signal amplitude: full signal amplitude in the main lobe of the window function & severely reduced signal amplitude outside that main lobe.

That's exactly what these audio apodising filters are doing - retaining the audio amplitude from 20Hz -> 18KHz then ensuring that the audio signal is fully attenuated by Fs/2. So, effectively this is a windowing function.

This bit of info seemed to go full circle for me & turned on a light. Just thought that I should share it w/ all of you.
FWIW.