I apologize if this has been thought of and/or discussed.
I am doing some ltspice simulations and it occured to me that crossover distortion in a typical totem pole push pull output stage would make harmonic distortion worse at the smallest output power that departs class a-ish operation.
eg for 1mA (hypothetical) idle current across the devices it would make sense that harmonics would deteriorate the MOST somewhere 1mA to 2.x mA output current.
This makes very low level listening an extremely different experience than the typical half power or 80% output situation measurements
For my particular simulation, THD @ max usable power (100+W) is < 0.0005% but at 100mW its more than 10 times worse.
Any thoughts?
I am doing some ltspice simulations and it occured to me that crossover distortion in a typical totem pole push pull output stage would make harmonic distortion worse at the smallest output power that departs class a-ish operation.
eg for 1mA (hypothetical) idle current across the devices it would make sense that harmonics would deteriorate the MOST somewhere 1mA to 2.x mA output current.
This makes very low level listening an extremely different experience than the typical half power or 80% output situation measurements
For my particular simulation, THD @ max usable power (100+W) is < 0.0005% but at 100mW its more than 10 times worse.
Any thoughts?
That is a great question particularly now that there are so many amplifiers on the market with very low distortion numbers. https://audiosciencereview.com So you are looking at 10 times more than < 0.0005% distortion. So that's 0.005% distortion. Maybe have a look at this website and see what level distortion you can reliably hear and report back here with the results. https://www.audiocheck.net/testtones_thd.php There is also Douglas Self's book on amplifier design that breaks down the performance of each stage of an audio amplifier and measures the distortion performance for all of the popular architectures for each stage. He actually built and measured the performance of the different variations and discusses the results in detail in the book. Amplifiers work in a system that includes speakers. I found that only by buying many different brand and model drivers, and testing each, it was possible to find drivers that had distortion levels as low as 0.1% at maybe one Watt or less input power. You might find it interesting to measure the distortion produced by your loudspeakers.
Last edited:
The "total" in "Audibility of thd" does not really make much sense in this case. It hugely depends on the number of "h".
low even order harmonics (e.g. H2) are very different from high order uneven harmonics. Also depends on the source material you listen to. And the relation of different harmonics.
low even order harmonics (e.g. H2) are very different from high order uneven harmonics. Also depends on the source material you listen to. And the relation of different harmonics.
I am not a stranger to D. Self's book, I am working based on that for much of the experimentation. 🙂
Crossover distortion however, being a narrow-time wide bandwidth problem is easy to distinguish compared to lets say 2nd harmonic only, as it has been already talked about (ad nauseam) across space and time 🙂
Funny thing, I am experimenting with a CFP output stage using 2sc5200Ns and I have to say the Vq I measured for minimum thd at the output stage roughly agrees with D. Self "Table 10.3. CFP output: data for 8 U, load and CFP o/p stage"
To be clear, I dont think 0.0005% THD is neither attainable in practice nor worthy of pursuit. blind a-b tests would not show that anyone can hear it is better than 0.005%. Is it more pleasant and liked by more though?
You are running into a S+N/N problem.
THD+N of 0.0005 is about 105dB and THD+N of 0.005 is about 85dB S+N/N
100W is about 28v and 100mw is about 900mv or 30dB lower.
Duke
THD+N of 0.0005 is about 105dB and THD+N of 0.005 is about 85dB S+N/N
100W is about 28v and 100mw is about 900mv or 30dB lower.
Duke
As you have a spice model it would be relatively easy to do a fixed time step simulation and convert the output file into a .wav that could be listened to. A script in Octave works well. Then you could make several with different levels of crossover distortion to find the threshold of audibility. Everybody has an opinion, but very few people have test results to share.
As well as the frequency, sound pressure levels and background/environment noise.
The only sense it makes, is that it is presumed that lower equals better, not knowing well where the threshold is.
In the end that's the equivalent of looking a 32k 50" display screen at 5 meters (~ 15 feet).
The interesting part is that the majority of people (incl purists) would find it crazy when it comes down to TV screens.
But for audio, people believe the resolution has zero limits it seems.
I already said that thd < 0.00... is not worthy of pursuit. 🙂 There is a well defined auditory limit, just not a pleasantly computed one. Auditory masking .
There is this excellent paper : https://royalsocietypublishing.org/doi/full/10.1098/rstb.2007.215, which as usual focuses on voice only, but if there are any hard limits/thresholds to be derived from this, they should be too strict for general audio, since voice is what the ear is really good at.
Long story short, I think that for typical listening levels the typically accepted numbers for harmonics (something with two zeros, 0.00x%) are an order of magnitude better than what the ear can process.
As far as high SPL, well we all know clubs and concerts where 10%THD at 10kWs makes the ears buzz for hours after the event, and that doesn't make people scream and run away with their hands up.
Low SPL, well, there, I think, it is much more difficult to reason about, especially in a living room.
There is this excellent paper : https://royalsocietypublishing.org/doi/full/10.1098/rstb.2007.215, which as usual focuses on voice only, but if there are any hard limits/thresholds to be derived from this, they should be too strict for general audio, since voice is what the ear is really good at.
Long story short, I think that for typical listening levels the typically accepted numbers for harmonics (something with two zeros, 0.00x%) are an order of magnitude better than what the ear can process.
As far as high SPL, well we all know clubs and concerts where 10%THD at 10kWs makes the ears buzz for hours after the event, and that doesn't make people scream and run away with their hands up.
Low SPL, well, there, I think, it is much more difficult to reason about, especially in a living room.
Consider that the glitch is already averaged over a whole cycle. You're picking it up as harmonic distortion only because the input stimulous is a cyclic wave. For a music signal, the glitch would be less predictable and may manifest as some kind of low-level 'grit' or 'fuzz'. It might even seem to "add detail".
You could try to capture the glitch on a scope and play with the topology to see how to minimise it. AFAICT one of the differences between a classic BJT push-pull and a MOSFET based one is that MOSFETs tend to leak a bit of current in the "off" state, resulting in a smoother gain curve.
Spice does not model real world switching issues like charge storage and beta reduction well, so the crossover zone is a lot more complex than you might think
Since
https://drive.google.com/file/d/1p0yeT5az6MWGyKP9z3kzy-rXFvbHNTmv/view?usp=sharing
This dual mono file has one original-ish channel from a flac file, and the other one is after 5hours of simulating the same channel passed through my awesome®️™️ amp simulation, scaled to the original as well as I could.
BTW, There is nothing ideal about this simulation, including some trace capacitance and inductance where I thought it might make a difference (it does, at the right spot of course), as well as some coupling to make establishing stability so much more fun.
I could share the file, or you could just insert a generic blameless amp in its place. I dont think it matters
Which channel sounds right, and what is left?
Here is a quick test:
https://drive.google.com/file/d/1p0yeT5az6MWGyKP9z3kzy-rXFvbHNTmv/view?usp=sharing
This dual mono file has one original-ish channel from a flac file, and the other one is after 5hours of simulating the same channel passed through my awesome®️™️ amp simulation, scaled to the original as well as I could.
BTW, There is nothing ideal about this simulation, including some trace capacitance and inductance where I thought it might make a difference (it does, at the right spot of course), as well as some coupling to make establishing stability so much more fun.
I could share the file, or you could just insert a generic blameless amp in its place. I dont think it matters
Which channel sounds right, and what is left?
Sean Olive on THD:
Dr. Earl Geddes on THD:
This is precisely where the signal-based distortion metrics fail. In our next paper we will show that .01% THD of one type of nonlinear system can be perceived as unacceptable while 10% THD in another example is perceived as inaudible. Even one of these simple examples is sufficient to invalidate THD as a viable metric for discussion of the perception of distortion. Furthermore, 1% THD is not at all the same as 1% IM, but we will show that neither correlates with subjective perception.
http://www.gedlee.com/Papers/Distortion_AES_I.pdf
Dr. Earl Geddes on THD:
This is precisely where the signal-based distortion metrics fail. In our next paper we will show that .01% THD of one type of nonlinear system can be perceived as unacceptable while 10% THD in another example is perceived as inaudible. Even one of these simple examples is sufficient to invalidate THD as a viable metric for discussion of the perception of distortion. Furthermore, 1% THD is not at all the same as 1% IM, but we will show that neither correlates with subjective perception.
http://www.gedlee.com/Papers/Distortion_AES_I.pdf
i could not hear a difference... 😳
I liked the music a lot! even if it probably has lots of distorsion from a classic analog synth...
how loud did you turn up the "virtual volume knob"?
it should be turned down to maximize crossover distorsion ...