"Very true, so we should stick to a simple bandwitdh limited measurement of the output, resulting in simple number of xx uVrms noise."
In theory that is a good idea - the filter could easily be built, but I think there is no instrument that is accurate in the uV area on ACV. Simply because the signal needs to be amplified by many dB in order to make it measurable. and that adds the noise of the instruments amplifier to the signal.
In theory that is a good idea - the filter could easily be built, but I think there is no instrument that is accurate in the uV area on ACV. Simply because the signal needs to be amplified by many dB in order to make it measurable. and that adds the noise of the instruments amplifier to the signal.
Just step into noise calculation of OP amp circuit and check numbers,
you will notice that a sufficiently low noise measurement amp is easy going.
The signal offers a low impedance and there are multiple OP amps with approx 1nV/Hz. But already 3-4nV/Hz types will do the job.
The 3-4nV/Hz integrate up over the audio band to approx. 0.5uVrms.
Resistor noise easily can happen to dominate over the OP amp noise.
The filter is more difficult, but also possible.
You have to fade out the 400kHz carrier by -110db, but allow a bandwidth of 16kHz for the noise measurement.
You can use the circuit, which I posted for jitter measurement.
In fact the clue of this circuit is to magnify jitter of a PWM in the audio band.
Obviously it can also be used for noise measurements of a class D amp.
It is working fine, but is not at all touching the limits of professional measurement equioment. About 20db better should be still possible, by going for more clever and lower impedance filter networks and choosing better OP amps.
Well, with the shown primitive circuit the equivalent input noise is around 5 uVrms, which is better than the output noise of my class D amps so far.
( Assuming a power amp which can deliver 40Vrms the 5uVrms translate to -138db. 5uV would translate into 3 pico Watt @ 8R, I am not there yet. )
http://www.diyaudio.com/forums/class-d/206549-low-cost-jitter-measurement-117db.html
you will notice that a sufficiently low noise measurement amp is easy going.
The signal offers a low impedance and there are multiple OP amps with approx 1nV/Hz. But already 3-4nV/Hz types will do the job.
The 3-4nV/Hz integrate up over the audio band to approx. 0.5uVrms.
Resistor noise easily can happen to dominate over the OP amp noise.
The filter is more difficult, but also possible.
You have to fade out the 400kHz carrier by -110db, but allow a bandwidth of 16kHz for the noise measurement.
You can use the circuit, which I posted for jitter measurement.
In fact the clue of this circuit is to magnify jitter of a PWM in the audio band.
Obviously it can also be used for noise measurements of a class D amp.
It is working fine, but is not at all touching the limits of professional measurement equioment. About 20db better should be still possible, by going for more clever and lower impedance filter networks and choosing better OP amps.
Well, with the shown primitive circuit the equivalent input noise is around 5 uVrms, which is better than the output noise of my class D amps so far.
( Assuming a power amp which can deliver 40Vrms the 5uVrms translate to -138db. 5uV would translate into 3 pico Watt @ 8R, I am not there yet. )
http://www.diyaudio.com/forums/class-d/206549-low-cost-jitter-measurement-117db.html
My experience with class D is limited to a 2050 chip Tripath amp...when compared to my class A F5 it has a lot less "air". A lack of treble extension for sure. But it does many things well..especially for the price. The tripath amp, while not current tech, was far better than the Rotel integrated it replaced. Soundstage is big though lacks a bit of depth. Bass is big but not that precise. I have been listening to just the Pass amp lately, because the tripath amp is sensitive to some sort of RFI interference caused by my furnace switching on and off. But summer has been brutal with the heat coming off of the big Class A amp and no AC.
Although the TK2050 chip was rated 50W it clips far earlier and seems to have less power than my 25 watt F5. And when it clips it is nasty. But that little amp is killer for less than $150.
Although the TK2050 chip was rated 50W it clips far earlier and seems to have less power than my 25 watt F5. And when it clips it is nasty. But that little amp is killer for less than $150.
This discussion is going nowhere until we define "air".
E.g higher notes decaying faster in amp A than in amp B?
And how can you be sure that amp B is closer to the source (recorded material)? Perhaps that faster decay is just an indication that amp A is faster and more accurate.
If I may take the liberty to quote posts from a different thread, here's a discussion about the F5X (an improvement - at least objectively - on the original F5):
E.g higher notes decaying faster in amp A than in amp B?
And how can you be sure that amp B is closer to the source (recorded material)? Perhaps that faster decay is just an indication that amp A is faster and more accurate.
If I may take the liberty to quote posts from a different thread, here's a discussion about the F5X (an improvement - at least objectively - on the original F5):
Last edited:
Choco, what is your preferred oscillator and triangle generator for your designs?
Originally Posted by ramallo
Is a XA30.5. I compared the F5X against a regular F5 too.
My "first impressions": I love the open sound of the F5X, is crystal clear, but sometimes is a bit strident in comparison against the XA, also have a bit less of low level resolution (Subjetive), I'm using the song "Ring Dem Bells" from Cotton Club Soundtrack, the bells dissapearing faster on the F5X than the XA. The soundprint is similar to the regular F5, but the F5X have better bass (More controled), and more relaxed performance.
This was done with the F5X at 1.5A, I'll repeat at 2A (When weather permitting).
My regular speakers are 8 Ohm (Two way, 96dB 1W/m) (DIY), too I did the test with a pair of Quad ESL63
When you guys compare amps is it a "memory" test or a blind ABX?
Is a XA30.5. I compared the F5X against a regular F5 too.
My "first impressions": I love the open sound of the F5X, is crystal clear, but sometimes is a bit strident in comparison against the XA, also have a bit less of low level resolution (Subjetive), I'm using the song "Ring Dem Bells" from Cotton Club Soundtrack, the bells dissapearing faster on the F5X than the XA. The soundprint is similar to the regular F5, but the F5X have better bass (More controled), and more relaxed performance.
This was done with the F5X at 1.5A, I'll repeat at 2A (When weather permitting).
My regular speakers are 8 Ohm (Two way, 96dB 1W/m) (DIY), too I did the test with a pair of Quad ESL63
When you guys compare amps is it a "memory" test or a blind ABX?
Last edited:
Simple analog solutions with a hysteresis comparator + integrator can perform very well, if done with the right comparator, layout and rail decoupling.
In SystemD MD I designed discrete comparators, but that's not mandatory.
SystemD 2k4 is self oscillating with integrated comparators and the overall noise is dominated by the symmetric input buffer gain stage.
Without the symmetric input buffer the power amp itself delivers an output noise of approx. 30uVrms.
There may be good instruments (like your schematic), but my LeCroy oscilloscopes or my Flukes w/o such pre-amplification would not show uVrms.
"You have to fade out the 400kHz carrier by -110db, but allow a bandwidth of 16kHz for the noise measurement."
In total with the LC damping that gives you >150dB - fair enough!
"You have to fade out the 400kHz carrier by -110db, but allow a bandwidth of 16kHz for the noise measurement."
In total with the LC damping that gives you >150dB - fair enough!
..Karl....strange guy... let him reckon.
Same measurements in my experience do not ensure same sound.
At least I do not know the right measurements so far.
Hi Tom,
correct.
With standard equipment for SMPS design you won't see it.
But it is easy to build this magnifying glass for noise.
Formerly I had a similar thing with just 1st order bandwidth limitation for my analog designs... But with class D and my curiousity to examine the phase noise of the triangle generator or PWM directly - I had to go the next step.
correct.
With standard equipment for SMPS design you won't see it.
But it is easy to build this magnifying glass for noise.
Formerly I had a similar thing with just 1st order bandwidth limitation for my analog designs... But with class D and my curiousity to examine the phase noise of the triangle generator or PWM directly - I had to go the next step.
Choco have you tried the following noise amplifier idea:
Take the amps output signal, amplify by factor 1000 or the like. Call it Signal A.
Put A through a specific highpass, that a) has almost zero damping and zero phase at fs, and b), good damping > 35dB at 20kHz (with no req. for phase in the stopband). Call this Signal B. Then invert Signal B (or A) and add both together. Signal at fs will cancel almost perfectly, 20 - 20k will remain mostly unchanged. Push this sum signal through a convenient lowpass to further suppress remains of fs.
Take the amps output signal, amplify by factor 1000 or the like. Call it Signal A.
Put A through a specific highpass, that a) has almost zero damping and zero phase at fs, and b), good damping > 35dB at 20kHz (with no req. for phase in the stopband). Call this Signal B. Then invert Signal B (or A) and add both together. Signal at fs will cancel almost perfectly, 20 - 20k will remain mostly unchanged. Push this sum signal through a convenient lowpass to further suppress remains of fs.
What are you actually trying to prove?
Try something useful like recording a mouse fart, you would make a name for yourself on Youtube
"Ummmhmmm Karl reckon if I can get through this one slow guy movie ummmm hmmm reckon on the next one I'll be neked with Halle in Monster's Ball mmmm hmmm reckon mmmm . Gonna check out her class d's see if there is any air in there... mmmmhmmm reckon, ummm hmm reckon.
Interesting. I haven't been happy with the "typical" op-amp based integrators. I don't know what state-of-the-art is for triangle generation, but I have been using switched CCS's to feed the integrator capacitor, and just using an op-amp to buffer the output. This triangle generator has improved the open loop THD in my 1MHz switcher. I have also considered experimenting with DDS triangle generators.
I also don't use a self-oscillating design. The main reason is that I like to use a single clock for the power amp and it's switching power supply, to eliminate a potential source of noise. A secondary reason is that can use the same oscillator designs I use for receiver LO's, which have vanishingly low phase noise. The real limitation on phase noise ends up being the downstream circuitry.
Last edited:
@ Tom:
...did not try your approach, but from theory it should work.
In real life it appears to me more complex than the trivial filtered gain stage.
Especially your first sentences Take the amps output signal, amplify by factor 1000 or the like. will need an amplifier with high capabilties regarding output voltage.
The carrier residuals of normal class D amps are ranging between 300mVp...2Vp. I dare to say that I could design a gain stage which can amplify this accurately by factor 1000 without clipping , but for me this is not a trivial one day activity.
@ Defiant:
Using specific low noise switched CCs for charging the integrators worked OK for me in one of my older designs. The advantage of this is that you can get away with average components. Using an OP integrator needs a fast OP amp for good results.
In system D MD I used the discrete comparators, but used a LT1363 for the integrator and got pretty fine results.
Take care with focussing on the open loop THD in class D amps.
Closing the feedback loop is adding a significant distortion mechanism to the modulator (described in the beginning of the MD thread).
Last but not least, even a perfect triangle is not the best. The optimum is close to the perfect triangle, because you can use small intensionally controlled deviations from the triangle to compensate distortions of the switching stage.
...did not try your approach, but from theory it should work.
In real life it appears to me more complex than the trivial filtered gain stage.
Especially your first sentences Take the amps output signal, amplify by factor 1000 or the like. will need an amplifier with high capabilties regarding output voltage.
The carrier residuals of normal class D amps are ranging between 300mVp...2Vp. I dare to say that I could design a gain stage which can amplify this accurately by factor 1000 without clipping , but for me this is not a trivial one day activity.
@ Defiant:
Using specific low noise switched CCs for charging the integrators worked OK for me in one of my older designs. The advantage of this is that you can get away with average components. Using an OP integrator needs a fast OP amp for good results.
In system D MD I used the discrete comparators, but used a LT1363 for the integrator and got pretty fine results.
Take care with focussing on the open loop THD in class D amps.
Closing the feedback loop is adding a significant distortion mechanism to the modulator (described in the beginning of the MD thread).
Last but not least, even a perfect triangle is not the best. The optimum is close to the perfect triangle, because you can use small intensionally controlled deviations from the triangle to compensate distortions of the switching stage.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.