Sauermann Amplifier

[PeterSt]

Many amplifiers do generate internal noise that is easily audible if you are direct driving horn compression drivers in a multiamped system.

Haha Greg, you seem to know more about my system than I myself. So, all still 100% correct ...
(and still no noise)

Anyway, very nice reasoning about the "attenuation" in the PCM1704. Maybe we should talk adding / subtracting currents, but okay.

Without analysers at hand, or pictures of it, all is theory of course. Or at least maybe it is for you. I mean (and for example only), just try to pull a plot of the noise line from an amp from somewhere (internet) that is 100% straight (up to say 96KHz). So there it starts. Btw, I think I already said that the Sauermann does this. Same for the GainClone.

That is one part only. A next is your attenuator, or preamp for the worst "solution".

Along with this I can only try to tell you (listening would be so much better ) that this noise line tells all. When that turns into something uneven (lumps, bumps, slopes up / down) the music is killed. It is just a measure for that. And not to forget, I worked for months on it until I had to give up (if only the PCM1704 could be voltage regulated).

It is a bit more difficult to relate this to the "analogue attenuation anyway" theory you applied so nicely, but let's say I don't see that happening.

It's just that by careful design BB keeps any filtering to a higher frequency than we will ever worry about.

That neither, but that's another story (what "filter" ... where ?).

So for example, when I attenuate digitally with 141dB, I have the (btw 16/44.1 !) signal just sticking out the noise, and this noise does not look differerent from no signal being there at all. Notice though that this could be the opposite situation of "attenuating" and I never really wondered what comes first in the 1704 (full signal which is divided, or no signal which is added, but the bi-polar etc. stuff should tell a few things).

Now, with my mentioned -21dBFS signal, the noise line still does not change, but, a few harmonics are there of course (that's just the THD from the chip, and losing bits if you like).
Only when I attenuate less than 21dBFS the noise line starts to rise linearly which in my case will be caused by the means of (Arc Prediction) filtering. But, this noise is still completely "horizontal" (straight). This latter only applies to the audio band, or more when higher resolution is in order (meaning : after fs/2 the noise level rolls off). Well, look here :

To keep in mind : The FFT depth used here (and further pictures) makes the noise look 20dB lower than realilty. So it's here at -140dB really.
Anyway, this is what I call a straight noise line. Measure this behind my GainClones and nothing changes except for the gain which is added plus a small 3rd harmonic (sticking out with 9dB IIRC).

This is 1000Hz at -0dBFS signal at -3dbFS (16/48 Arc Prediction upsampled to 24/192 (and *not* to 24/768 which the NOS1 can digest). All still straight.
(notice this is up to 10KHz (just grabbed an existing picture from somewhere)).

Here you see what I talked about regarding the fs/2 thing. The 20 dB added must be from my filtering (I see this as random HD resulting in white noise) with possible reactance in the gain stage.
Each dB of attenuation will drop the noise floor linearly, until it gets under the inherent noise (this is what I talked about in my earlier post).

Here you see the proof of the fs/2 thing, since this is a 24/96 signal.


Ok. As you can see, noise lines are not sooo straight as I implied they should be, but this is software impeded. To my ideas this can't harm because it "goes on with the flow" and further more it is as straight in the audio band intended (fs/2).
Now add any attenuation means (of which I don't have pictures, or at least not at hand). On of the means I obviously tried was your type of attenuator (I won't name the brand to not discredit well respected people), relais regulated. It was the worst of all I tried. Look at the first picture again, but now imagine lumps all over like you see beyond my fs/2 pictures. But then in the audio band of course. Spikes all over the place, and of course varying per chosen attenuation (btw, digitally attenuating with the ladder DAC shows different harmonics per attenuation level just the same - just saying, but in normal "expected" fashion).
Maybe you can dig up plots from your attenuator, or otherwise go out on Google and find nice straight noise lines which are not 100 times smoothened (look at the thickness in my plots).

Anyway, this is measuring. But before that comes listening. Ad then what's audible is visble at measuring. It just is so. In all situations I tried (up to one-value fixed resistors in all kind of setups, yes, also right in front of the amps).
Any means of transformer (I/V) ? it just can't work (with a TVC ahead as a rough one). I went as far as buying a most official LDR for USD 450 or so, and then to think this was for two "ch" only (so not even able to work in differential mode which would add another 200 or so).

In my case you can also say that I chose the wrong chips of course. But in the end not, because doing it digitally is just the most okay and really creates the best sound by far.
And not to forget, in the 1704 situation 90% of the work in this context goes to the I/V itself, which is a pain ... (but very much "volume" related).

Aaanywayy ... no noise here with dual GainClones and compression drivers. Haha.
Oh, this was about the Sauermann. Well, as said, same story. Same noise line too (and I say this is RARE).

Thanks for sparring a bit.
Peter

[Flecko]

Nice to read Peter and Greg.

Peter, you told your system has "no noise". Does that mean you cannot hear anything even if you put your ear directly at the tweeter? With 110db/W speakers this would be an incredible achievement. I finally got rid of my preamp (...I must find it out myself) and now also the noise is very low but not unhearable. I can hear a noise comming from my tweeter if my ear is 5cm away from it and if just my amplifier is turned on. I can hear a noise up to 15cm away from the tweeter if my dac is turned on too. As it seems to be very important to get the noise down in the system, it would be nice to have a to do list to reduce the noise. Maybe you or Paul could start a thread about this?

One thing disproving one argument of my own: The increase of what I thought was distortion at very low power, smaler than 1W, could be very likely caused by the noise in the THD+N plot. The pure distortion should be smaler with lower output power. So no better performance with low sensitivity speakers here.
The noise is at least less hearable. As the hearable noise should be normaly come from the DAC/Preamp section, the amplifier should be matched by choosing the right amount of gain to fit to the speakers and the DAC.

[CoenP]

Noise, noise, noise.

Great posts

The whole xxhe pc playback is in the end about noise. Are you able to see the effect of different xx settings? Is this plot made with v07?

So this is where the Byebees come in, and everything else. Even vacuum tubes  , at least on my side!

Regards, Coen

[PeterSt]

Does that mean you cannot hear anything even if you put your ear directly at the tweeter? With 110db/W speakers this would be an incredible achievement.

Almost correct Adrian;
Speakers are 115dB/W @1m.
So, gain of the amps is around 21dB and the base is some 8uVRMS coming from the DAC (idle but On of course).

It now depends a bit how my interlinks ar routed whether you can hear totally nothing or just that little bit. Coen here is witness of how he could hear that little bit the other day (ear completely in the horn) and maybe he can also tell you how my interlinks (SE) went about. I forgot. Haha.


Edit :
To be honest, this picture shows the output noise with balanced interlinks; with SE (RCA) the noise is ~17uVRMS (say 6dB of difference). Since this is what I use (SE), it is this what I will hear (after the amplification of the amps of course).

[PeterSt]

The whole xxhe pc playback is in the end about noise. Are you able to see the effect of different xx settings? Is this plot made with v07?

No, this is from much longer ago.

The effect sure can be seen but this is merely in the amount of upsampling. So remember, those pictures were about 16/48 upsampled to 24/192 (32/192 actually) and upsampling to 24/768 changes the picture more or less dramatically on where the harmonics fall. So, think about (false) harmonics being pretty much the same for interval (like each 1000 Hz distance for a 1000HZ signal) but some are higher, some are lower (like you see the 3rd is higher in the 2nd picture above). For "nasty" frequencies and nasty harmonics this shifts to outside the audible frequencies (say beyond 20KHz) but this only happens at this 16x upsampling (not at 8x yet).

So, this is not really XXHE version related, but let's say that this is only not the case because I never changed a thing about the filtering (Arc Prediction) means from when it was introduced. It could (be improved) though.

Regards,
Peter

[Flecko]

Does that mean you cannot hear anything even if you put your ear directly at the tweeter? With 110db/W speakers this would be an incredible achievement.

Almost correct Adrian;
Speakers are 115dB/W @1m.

 

[gsbrva]

It's true that I don't know Peter's system.  I was using examples that I thought most readers could relate to.  My reference to 115db 1watt goes back to Altec(and other) horn drivers I've used.   I'm sure Peter has talked about his speakers, but I haven't read that yet.  As far as chip amps go, I heard a few and repaired a few.  I'm neutral on that topic.  They are what they are and seem to be extremely good for the effort and parts cost.  I have never tried running a chip amp directly into a high efficiency horn driver.  It's interesting that the GainClones are so quiet.

Nice plots by the way, thanks for sharing that stuff 

Maybe I should take this to another thread, because it's going far off topic.  I am interested in the Sauermann, honest!  Sorry for the hijack......

Read datasheets at night to cure insomnia......., I'm actually quite familiar with the internal diagram and function of the PCM1704 via the Japanese application notes.  It's true that the differential bit switches are operating more or less as current switch only, but the impedance of the rest of the R2R ladder is certainly not zero and voltage is swinging(making capacitive reactances matter).  Hence my statement about it being filtered.  Also, if you have succeeded in making a perfect current sink for the I/V conversion, then my hat is off to you Peter! 

I can't think how the I/V can be made a perfect current sink, so I'll assume a small voltage swing on the current output as well.  Therefore, the summed bit current output is limited in the rise time.  This is a filter, it's just so far out of band that it doesn't matter.  Sorry to be argumentative on that point.

It's all just reactances and no different than what happens when you insert an analog attenuator followed by an interconnect, amplifier, etc. 

If a simple voltage divider sounds poor compared to digital attenuation into the input of the GainClones, that says more to me about the input impedance and reactances of the chip amp than about the divider resistors.  Another thing to consider is the sonic effect of the output levels on the I/V.  Different levels might sound different and internally on the DAC chip. From what I can see of PCM1704 there may be sonic advantages to running at a sweet spot in the ladder with regards to major bit crossings etc.  I know the colinear arrangement fixes the zero crossing, but other major crossings exist. 

I don't have a clue what modern commercially available attenuators are out there.  I build my own stuff and also have some lab grade General Radio decades (T and H networks).  I've listened to TVC at a friends house and they just sound like transformers to me.  A bit of dynamics and air lost.  Not my choice.  They did seem to help filter the output a Sabre32 (Buffalo II) as a plus.  BTW, as far as I can tell, this old 16bit Phillips chip beats the Buffalo II when using XX arc prediction.

I'm using transformer I/V right now because I'm lazy and it's very safe for a 4 year old girl to play ABBA and dance   The gain is limited so that 0db cannot arc the Quad ESL63's.  Trying to teach her how to run XX.........

Regardless of my current system, I'm completely in favor of active device I/V.

Back to digital attenuation.......  How can we use the same bits for two different jobs?  If you attenuate 48db, are you not using all 8 extra bits below 16?  At the same time, don't you need those bits to interpolate 16x FS in the NOS1?

48db is a lot more attenuation that anyone needs.  But realistically, 24db is not.  If you use -24db attenuation (gives up 4 lowest bits (6db per bit) and also use 16x oversample, where are the discrete data points to create an arc?  If two words are 1 LSB apart, then having just 4 bits (16 levels) does not allow an arc (only linear), or am I missing something?  Isn't 16 integer points plotted over 16 samples a straight line?

Also, it seems practical experience shows that your digital attenuation sounds musical (many NOS1 users).  Maybe we don't need so many extra bits to make the arc prediction idea work?  I know that on my humble 16bit dac, arc prediction at 4FS seems to work surprisingly well.

More food for thought.  Has anyone analyzed how often two adjacent words in redbook pcm happen to be within a few LSB (least significant bits) of each other.  If that doesn't happen often, it would help explain how arc prediction can work well with less than ideal bit depth (as when using 24db of attenuation).

Yes, I am building a PCM1704 dac (as promised), just very busy right now!

Thanks for the entertaining conversation.  I hope I'm not irritating anyone and again sorry for the topic hijack (I'll move if it continues).  My mind tends to wander around with this stuff.  It's all so interconnected.

Greg


 

[PeterSt]

Hi Greg - Two small remarks for now, and I'll be back with some others probably tomorrow :

1. Gerd Sauermann is as interested in the NOS1 than others are hopefully in his amplifiers. So I don't think we need to worry much about hijacking the topic. It is all about a perfect match (of our both ideas) here - by accident (or not).

2. You are 100% correct on not being able to use the same bits for the different jobs (you know quite some stuff here). I didn't want to dive into it too deep because there's more. But this is why I mentioned the 16 bit (!) signal still sticking out its neck at -141dBFS, which includes the digital attenuation *and* 16x upsampling (onviously). No dither used ...

More later. And thanks for some great posts ...
Peter


PS: I have that Japanese (extended) datasheet too. Can't read it all though. Haha.

[PeterSt]

Ok, a couple of more responses ...

Therefore, the summed bit current output is limited in the rise time.  This is a filter, it's just so far out of band that it doesn't matter.

In band (sort of) it goes like this :

IIRC this is something like 27KHz Dirac pulses. This is measured from the NOS1 output. This is 16/44.1 upsampled to 24/176.4 which gives the pulses some width (of 4 samples instead of 1). Upsample to 705.6 makes them wider again, which obviously is for the better. So, supposed these kind of "transients" are in the music data, then the upsampling (better : filtering) makes them smooth enough not to harm the remainder of the chain. Or at least softens it.

Since this topic is still about the Sauermann amplifier ...
I don't recall I measured this at the output of the GainClones, but I did measure the Sauermann. It shows exactly the same picture as the DAC output shows (but with more output voltage of course). So, all still "perfect" ... (but make your speakers follow this, or otherwise we did wrong ).

If a simple voltage divider sounds poor compared to digital attenuation into the input of the GainClones, that says more to me about the input impedance and reactances of the chip amp than about the divider resistors.

Ok, let's be careful that this is not too much theory only. I mean, yes, okay, true perhaps, but sadly all I hear I measure at the output of the DAC already. Now, of course, we can say that the analyser suffers from the same, but if so I rather think towards the inavoidability of this all. For example, I myself use the most poor (well, $1) stuff on cabling and I do this because I can't expect others to mimic my super duper noise rejecting etc. (if true at all) own stuff. So I approach this the other way around (and even ship the NOS1 with the same poor cables).
Anyway, I can't agree.

They did seem to help filter the output a Sabre32 (Buffalo II) as a plus.

$300++ interlinks also filter nicely. And is also no solution.
Solve it at the source.
(I'm sure you think the same, but I want to emphasize how in 100% of cases things are solved the wrong way (for the worse). Ok, make that 99.99%.

I'm using transformer I/V right now because I'm lazy

This is for fun and may not show much common sense ... :
What I think is that these kind of "gain" solutions can not exist. Theories may be nice, but practice is that there is no free lunch and whatever is "gained" is taken from somewhere else. So, (I) think like the gain in the audio band is taken from outside of it as a rough example. Or, that the gain for a 80Hz frequency is gained from the reduction of 15KHz. Something like that, and it can be seen in plots.
As I said, the worse.

[Scroobius]

Hi Greg - good to see you on the forum and joining in with the spirit of things. As my name is mentioned above I thought I had better add my few pennies worth.

First of all let me tell you that I have achieved a level of noise that is vanishingly small and that I did not think possible. My speakers are 98db OK not as sensitive as Peters but still pretty sensitive. How much noise do I get now? well I have to place my ear right inside the cone of the speaker and right up against the dust cap to hear anything at all - even then it has to be very quiet in the room to hear anything. Noise levels are tiny - most people just cannot hear any noise at all. Now that is quiet and way better than I have ever achieved before (and I do not have any balanced connections). And boy did it make a difference to sound quality. As Peter has pointed out elsewhere noise does not just sit harmlessly under the signal it modulates it. So what most people think as innocuous noise is actually very very harmful because it imposes its character on the signal - a very bad thing as I now know.

All my systems generated noise in the past some more than others but all produced hugely more than my system does now. And how did I get the noise level down? - by sorting out grounding and earthing (they are separate) properly and in a way that is just not possible with any shop built system IMHO. So the totally unforeseen advantage about building Gainclone amps is that (provided you know how your other equipment is built - and provided it is built with the care and attention paid to NOS1) you can sort the grounding arrangements to almost totally eliminate noise which is in turn mostly caused by earth loops and RF. After all GC's are based on cheap mass produced chips so they must produce more noise than high quality purpose built hi fi but I hear virtually no noise at all from my humble GC's. So where does most of the noise come from? not the amp from what I have heard.

Of course with the incredible NOS1 at the front end changes in sound quality are very obvious.

All the best

Paul

[PeterSt]

One small additional remark :

After all GC's are based on cheap mass produced chips so they must produce more noise than high quality purpose built hi fi but I hear virtually no noise at all from my humble GC's.

Although several GC chips exist and yours will not be the same as mine, the little trick is a sort of other way around :
With this one-chip design, there's virtually nothing else that can produce noise. There's maybe one feedback resistor only and that's all. The remainder is in the chip (comes from the chip). No outboard voltage regulator impeding noise stuff and such.

Your beautiful "hi fi" component may have dozens or more noise creating parts in the signal path. These too are mass produced cheap thingies (in general). Now, choosing those components for the best performance (on THD+N) is that other trick. But if I tell that to an EE, he tells me I must be nuts because any spec will do hence will produce inaudible noise ...

Peter

[gsbrva]

Hi Paul,  thanks for the greeting.  Congrats on getting your system quiet.  I couldn't agree more that it's essential to getting good sound.

Ok, a couple of more responses ...

In band (sort of) it goes like this :

IIRC this is something like 27KHz Dirac pulses. This is measured from the NOS1 output. This is 16/44.1 upsampled to 24/176.4

You can kinda see my point about filtering in the dac chip when looking at the rise time of the transients.  Are the peaks full scale?  It looks better than I expected.  Your I/V stage is certainly doing well!!

Ok, let's be careful that this is not too much theory only. I mean, yes, okay, true perhaps, but sadly all I hear I measure at the output of the DAC already. Now, of course, we can say that the analyser suffers from the same, but if so I rather think towards the inavoidability of this all.
Anyway, I can't agree.

LOL, me either.  We may have to agree to disagree

Ultimately, all solutions either sound good or they don't, right?  Obviously, exclusive digital volume control is working for many people.  However, I'm having a hard time wrapping my mind around the idea that it is best in all cases.  I hate the idea of using up much bit depth for attenuation, since it's a function that I believe is done very well in the analog domain (I know that is the point, lol).

Various combinations of speakers and amplifiers can have such drastically different gains that I wonder how a fixed output DAC could correctly interface with them all.  I personally own speakers that are nearly 30db apart in apparent sensitivity (horns and electrostats).  At what amount of digital attenuation would you be better off just padding the analog signal?

When I finish my 24bit dac, I'll give it a try and see, maybe I'm wrong!   

$300++ interlinks also filter nicely. And is also no solution.
Solve it at the source.
(I'm sure you think the same, but I want to emphasize how in 100% of cases things are solved the wrong way (for the worse). Ok, make that 99.99%.

Ha ha, not touching that one. 
javascript:void(0);

I'm using transformer I/V right now because I'm lazy

This is for fun and may not show much common sense ... :
What I think is that these kind of "gain" solutions can not exist. Theories may be nice, but practice is that there is no free lunch and whatever is "gained" is taken from somewhere else. So, (I) think like the gain in the audio band is taken from outside of it as a rough example. Or, that the gain for a 80Hz frequency is gained from the reduction of 15KHz. Something like that, and it can be seen in plots.
As I said, the worse.

Yes, I agree that nothing is free.   Except in the case of my little transformer I/V experiment.  These Altec A256C amplifiers happen to have low Z input transformers as an integral part of the design (phase splitter and balanced feedback mixing).  It was just too easy to hook it all up.  The transformers are already there and I didn't add them in circuit.  Now you see why I said "lazy". 

The amps are old 1950's stuff.  I finally restored them to operation after 20 years on the shelf in my basement.  I felt like I had to use them a bit after such neglect.  I don't have pictures to post, but here's a link for a pair.  http://cafe995.daum.net/_c21_/bbs_search_read?grpid=fDQc&fldid=7foC&contentval=0009lzzzzzzzzzzzzzzzzzzzzzzzzz&nenc=y8McCQFicrtqJ1TisjO-_g00&fenc=&q=&nil_profile=cafetop&nil_menu=sch_updw

Using these amps is kinda like driving an ugly car that just keeps running.  I haven't had the heart to put them back into retirement.  Soon though!

Greg 

[manisandher]

One disadvantage of the digital volume control is that you must listen to all the noise all the time.  I think it's nice to have an analog gain control located as close as possible to the  final amplifier stage to manage the noise.  That way you can run your digital volume control up near full all the time.  This preserves the bit depth of the converter and still allows reserve gain for quiet recordings without the disadvantage of a high noise floor for normal recordings.

Hi Greg, I never used XX's built-in vol control for a long, long time. But (quite a while ago now) I did a comparison of the following three different attenuation techniques:
- XX's digital vol
- Audio Synthesis Pro Passion passive (Teflon insulated high purity silver conductors and precision bulk-foil Vishay resistors) with a very short IC to power amp
- Pass Labs X1 preamp (certainly not the last word in preamps but considered good value for the £5K I paid for it ~12 years ago)

I still have the Pro Passion and the X1 and could re-conduct the comparison if necessary. I can't find the thread where I posted my findings, but I do remember my ranking. To my utter dismay (as a strong analogue attenuation advocate up to that point) it was:
1st) XX digital vol control
2nd) Audio Synthesis Pro Passion
3rd) Pass Labs X1

IIRC, the Pro Passion was very close to XX's vol control, but lost some of the dynamics. The X1, even with it's pretty advanced vol control was waaaay behind the other two.

Mani.

[PeterSt]

I hate the idea of using up much bit depth for attenuation, since it's a function that I believe is done very well in the analog domain (I know that is the point, lol).

Greg, I wasn't even aware of that this indeed is the whole point inyour view. But - or because ... this is a completely different subject. So, to my serious ideas this is not done well at all in the analogue domain. At least it is so different that you might wonder how the both can be compared at all.
But let's not put this to the debate, and actually I never elaborated about it (although I asked the question many times in this forum with nobody ever coming up with an answer - let's stick to that ).

Various combinations of speakers and amplifiers can have such drastically different gains that I wonder how a fixed output DAC could correctly interface with them all.  I personally own speakers that are nearly 30db apart in apparent sensitivity (horns and electrostats).

You are very correct here. Still nature depicts (so to speak) that this turns out quite all right actually always. I mean, when you have e.g. 115dB speakers you don't use 700W Hypexes at the same time, or ?
But nothing says you really won't, although possibly it may tell that the setup isn't optimal when combined so indeed (outside that DAC connection I mean).

With my own genuine experience about the NOS1 and its customers I don't know of any single situation where people are in lack of gain, or have too much. However, it for sure is true that most will have more gain than they like - theoretically. I myself play in between something like -33dBFS and -12BFS, with the normal range between -33 and -21 or so; that I sometimes "need" -12 is because of very soft albums (Crime of the Century is always my example). These very soft albums are not the best to begin with of course, because they lack some 9dB of dynamic range, and that in the 16 bit domain (this is just poor engineering IMO). So counting those out, I am unnecessarily attenuating 21dB at least. But no way it harms or can ever cope with analogue attenuation (oh, I said something like that already ?). And as a matter of fact, there is also no way that I ever ever perceived degration per attenuation level or something, and this while we can fairly say that I am about everything which could make a small difference (be that software or DAC).
... And I also have to meet the first one who tells me such a thing ...

Btw, you won't know it I think, but the NOS1 is to be used without pre amplifier or other means of analogue attenuation. The text about this comes close to "forbidden".

Peter

[PeterSt]

You can kinda see my point about filtering in the dac chip when looking at the rise time of the transients.  Are the peaks full scale?  It looks better than I expected.  Your I/V stage is certainly doing well!!

Yes, full scale (2.25VRMS).

I can not only kinda see your point about the filtering, but sadly you will "lose" here. So yes, it officially is necessary, but no, it can't be done; It can't be done analoguely because you will or heavily influence the normal audio band, or it isn't effective at all. It can't be done digitally because that violates the whole design principle for starters (ok, that's a moot thing by itself of course) and next will kill the sound more than whatever analogue attenuation. This latter is not a moot thing at all, which makes the design theories also not much moot. Of course you can only know when you listen and compare, which in my (NOS1) case is the most easy by just applying the different filters in-software.

In between the lines : you can't guess how much this is Sauermann related ...
(but Gerd better tells that himself if he likes to)

Of course you could only refer to analogue filtering (because what you see in the picture is the result of digital filtering to begin with (Arc Prediction), but as I said : it can't work effectively unless you want a huge roll off of the high frequencies. And it is this you should not want, because it is this where something like the NOS1 excels.

Notice that the picture doesn't show music and it can only be a test signal. Still it is the resemblance of super high transients which exist in music 100 times more than you could ever expect. Not that a hit on a snare rim would incur for it, but at the micro level it does like it can be in women voices. Of course we could say that this "one sample" rise (time) is because the resolution (sampling rate) is too low to spread it over more, but in the end this doesn't matter because it is about the time the transient spreads over, and these transients (spreading over 1/44100 of a second) do exist. Not all at full scale, but thousands in a track, exceeding half of it easily. So yes, that should be filtered by itself, but this happens and this is why that one pulse spreads over 16 samples to begin with and why the higher upsample rate is important. Still it will alias, but this is beyond the audible range, and flattening it with normal (sinc) filtering will flatten all to nice sines. The best idea to turn violines into flutes, but also the reason why you (Greg) work with 1704's instead of some SDM (Sigma Delta Modulator).

Peter