I think it's important to also note that what really matters is the threshold of someone being able to hear the distortion--not just being able to measure distortion, as the OP doesn't make that distinction very clear.
In the case of THD, I assume that threshold of audibility numbers are found using headphones or loudspeakers playing either pure tones or perhaps tonal sequences that mimic some musical instrument tone, then they add some form of harmonic or modulation distortion on top of the generated test tone for detection of human threshold levels.
In most cases, I would think, these very low level distortion levels for finding hearing thresholds are much more revealing than typical music, in which the distortion levels would need to be much, much higher for humans to detect which tone has added distortion and which does not--like Keith Howard did a couple of decades ago with FMD: Red Shift: Doppler distortion in loudspeakers. I believe Mr. Howard has a web site where he posted some apps that do the job nicely.
Chris
In the case of THD, I assume that threshold of audibility numbers are found using headphones or loudspeakers playing either pure tones or perhaps tonal sequences that mimic some musical instrument tone, then they add some form of harmonic or modulation distortion on top of the generated test tone for detection of human threshold levels.
In most cases, I would think, these very low level distortion levels for finding hearing thresholds are much more revealing than typical music, in which the distortion levels would need to be much, much higher for humans to detect which tone has added distortion and which does not--like Keith Howard did a couple of decades ago with FMD: Red Shift: Doppler distortion in loudspeakers. I believe Mr. Howard has a web site where he posted some apps that do the job nicely.
Chris
Last edited:
Hi ! thank you very much for your valuable advice If we agree that in the end speakers have a much bigger impact on the overall sound the quality of the electronics becomes less critical ? in the sense that they should not have clear faults like too much noise and very high odd order distortions ?
it could be very well the case
Hi thanks a lot again for the very valuable input. Next time i will ask first. My idea was a little weird
I think that an headphone can be a good way to test a headphone amp sound quality. I have already some decent HPs and the idea was to select a good sounding headphone amp to be used also with a switch to power a low sensitivity power amp
An headphone amp can source 5 or 6 volts Way too much for a common power amp 4 or 5 Vgain in the power amp could be enough
Considering that i do not need more than 40-50W at the speakers this means 20Vrms at 8 ohm speakers terminals
Another solution could be a low power power amp with great current delivery to drive properly also low impedance speakers
i see your points But my main issue have always been preamps I see headphones like the stethoscope of a doctor
A very good heaphone is very revealing about what is upstream i.e. source and preamplification
I cannot check soundstage with an headphone but if the sound is good in the headphones is a very good start
Then it will be just a matter of provide some good clean current to the signal to move high efficiency speakers
very interesting point and i agree Moreover a very good system should sound "complete" even a low volume Like the bass range should be present like the mids and highs with no need to raise the volume at all
I have been exposed to small 2 ways ... they have no chance to provide a full range sound Something will be always missing
Hi and thank you sincerely for the very kind and interesting information
I have a problem My mind tells me that speakers set the overall performance But i like electronics so much more
I can tell a story Many years ago i bought a small T-amp quite trendy at that time On my medium efficiency bookshelf speakers it sounded not completely convincing Then i took it to a friend's home and he connected it to a pair of high efficiency speakers with mid and high horns
What a sound ! but the responsible for the great sound were the speakers not the amp
Since then i decided to look at low power amps and considering high efficiency speakers
Yes.
There are some things that you can do to recover the produced music tracks from mastering practices that always seeks to make the tracks sound louder (even though no one I know asked them to do this). These particular techniques work especially well for music produced originally before 1991 before multi-band compressors became widely available to the mastering people:
The Missing Octave(s) - Audacity Remastering to Restore Tracks
More recently, I've added the following capabilities using a multi-band upwards expander in conjunction with the above techniques for more recently produced music:
Dynamic Decompression?
If you have any interest in trying out these techniques, send a private conversation my way.
Chris
Last edited:
That may be true in theory, but a theoretically pure sine wave must also have infinite length in time. One consequence is that FFTs often use window functions to smoothly fade in and fade out the measured signal, so as to eliminate transient conditions at either end. Especially something like a small DC offset.
So it seems ironic that with extremely high resolution HD testing we work harder to filter out the most interesting parts when most of the distortion is likely to occur.
It sounds like you're saying that windowing removes the smallest signals near the crossover region. Windowing does not do that. You'll have many, many zero crossings included within the window. For example in a 1M FFT of a 1 kHz tone sampled at 44.1 kHz you'll have nearly 24000 zero crossings included within the window.
Also, you don't have to use an FFT to measure distortion, including IMD. You can do that with a swept oscillator, a mixer (analog multiplier), and an RMS detector. That's how the HP 3581A works, for example.
These days we use FFTs because sampling and data processing can be done at incredible speeds with very high fidelity. Memory is free so taking, say, a 1M-point FFT is no big deal. The impacts of the FFT length and window functions are well understood.
Tom
No, I meant time-wise, near the start when a flat-line input signal changes into a sine wave, and then back to nothing at the end. Whatever you do, it must create a low frequency offset. Counting just a short 180° burst, there's a huge DC offset because there's no negative half of the wave. Adding another 180 balances it out, but there's a lag.
36000° gives 100 cycles, but the situation is similar. The DC offset created by the transient at the start is never fully balanced out until an opposite 'blip' is included at the end. If we take a long FFT and mute the start and end with something like a Hanning window, we're prioritizing the middle and getting rid of information about how the amplifier behaves in response to the transient at the start.
Not necessarily. Synchonous FFT does not need a window (or a window called flat top which is the same) so no start-stop transients and fade in/fade out.
It just takes exactly one (or more) integer cycles.
No serious designer does it anymore with the issues you describe - even a $ 100 soundcard and REW does synchronous.
Jan
I fully agree with the statement and please let me share some thoughts on this.
IMO we have inherited that strong desire for 0.000...% THD from the era of magnetic recording technologies - tapes and phono.
In those days it was a common practice to measure THD at 1 kHz and IMD with 2 tones: one like 60 Hz and second of some kHz.
Despite good figures some devices sounded good while others did not, but why?
Magnetic technologies have inherent property: pickup device (magnetic head or MM/MC cartridge) output voltage is higher for high frequehcies.
Thus input circuits for head/cartridge preamps have to be very sensitive in low band and have to deal with relatively high level signals in HF area - RIAA curve is self-explaining. If this dilemma for input stage is not addressed properly by engineers the preamp may have a respectful figures for THD and IMD (when measured as described above) but have "dirty", unpleasant sound.
When the phenomena was discovered later a dual tone IMD measurements were introduced (e.g. 19+20 kHz), it effectively reveals this flaw giving huge IMD figure when input stage cannot cope with HF bursts.
But what we had before dual tone IMD measurement become widespread? The devices where engineers invested in linearity of input stages may exhibit moderately better THD (let's say 0.03% vs.0.1%) but dramatically better sound because real IMD (available with dual tone measurement only) was lover by orders!
This still to be true for power amps - old power transistors were slow, amplifiers were hugely corrected to ensure stability with NFB, thus NFB was shallow in HF area - again, we may have a good THD @ 1kHz and lots of "hidden" IMD, not revealed by old measurement methods. But when engineer invests in bandwidth and linearity - you can get much better sound with a humble improve in THD.
Well, world have changed. Now we have DACs with vanishingly low distortions and PAs hitting 100kHz power bandwidth.. For me, I see no reason to stick on THD measurement @ 1 kHz anymore, dual tone IMD test gives comprehensive linearity estimation.
Also I think I wouldn't try to "squeeze" better than 0.01% IMD (100 ppm) out of any circuit, BTW my hearing declines over age )))
The DC offset when a sine wave starts playing is a physical reality, not a measurement artifact. It's a true offset that the amplifier 'experiences', and various charge levels get adjusted, severely at first, before drifting asymptotally back to centre.
I'm not sure what you meant with synchronous FFTs or integer cycles. If multiple FFTs are spliced together sequentially, then that would be just another way to ignore "anomalous" or inconvenient results, by playing a continuous wave and cherrypicking the desired distortion after a few seconds or however long it takes to drift to a minimum value.
This partly boils down to 'DC' being loosely defined as practical ultra-low frequencies, rather than a theoretical construct.
Your mind tells you everything correctly. I just want to add that the perception of sound from the speaker system depends not only on nonlinear distortions, but also on linear distortions, and if you really want to improve the sound quality, you will have to monitor not only the THD but also the shape of the amplitude characteristic of the speaker system in the listening area.
I find this the most offensive distortion in the whole recording/playback chain. You don't need to have a particularly good hifi system to easily distinguish between a good recording and a poor one.
I would add, "and phase response/characteristic" for attaining higher fidelity (also here).
In my work with fully horn-loaded loudspeakers that can control their polars down to the room's Schroeder frequency (e.g., 100-200 Hz for home-sized listening rooms - like you see in the first link), I really couldn't believe the difference in subjective sound quality once the all-pass phase growth of the crossovers was compensated or removed. It was difficult to describe but really quite astounding when heard. The catch is that in order to hear this subjective performance the loudspeaker's output polars have to be controlled to the point where early reflections from just around the loudspeaker in room (including floor and ceiling bounce) do not overwhelm the direct arrivals from the loudspeaker itself. That basically rules out almost all direct radiators except perhaps really large dipoles, in favor of fully horn loaded, including and especially MEHs.
Chris
The phase characteristic is included in the concept of linear distortion.
I did not focus on it because it is a very slippery topic that needs to be entered into with a full understanding of what and why is being done.
I agree, but it is still the most commonly quoted figure, which is why I provide it.
I also provide the SMPTE IMD measurement (60 Hz + 7 kHz @ 4:1 ratio), 1 kHz + 5.5 kHz (1:1 ratio) as proposed by Siegfried Linkwitz, and 18+19 kHz @ 1:1.
The tests tease out different things. The SMPTE test supposedly teases out power supply issues. The 18+19 kHz test gives you a clue about the amount of loop gain available at 20 kHz, and the 1k+5.5k test, Linkwitz argues, correlates better with the perceived experience because the IMD products fall within the frequency range were the ear is the most sensitive.
I've argued for a while now that a multi-tone test is an even better test. The test signal is pretty close to music. I (and others) also nudged Amir at ASR to perform those tests, though I think he's reporting the findings incorrectly. The IMD 'grass' should be compared to the RMS sum of the test tones, not to the amplitude of each tone, so I think he's pessimistic by ~22 dB.
Here's an example from one of my amps:
Tom
