Why decrease in frequency, increase in distortion?

[Kokoriantz]

The LM3886 or TDA7293, get the distortion increased with decrease in frequency. See curve bellow. They are opamp type with dominant pole of less than 20hz. The lower is the frequency, the higher is the NFB, so it must continue to decrease.
Thermal feedback?

[srb1954]

Possibly thermal fluctuations across the semiconductor die causing thermal gradients affecting the balance of the input stage of the amplifier. As the operating frequency goes lower the thermal mass of the die is less effective at smoothing out the temperature variations within the die.

[EE_Audio]

Without seeing the test circuit, my guess is that it is just 1/f noise.

[unikeyname]

Do you mean the source of the noise is from parasitic LC?

[Kokoriantz]

The power is constant 30w, hence the SNR is constant for all frequencies. Note that it is measured without filter.

[Kokoriantz]

The more the low level transistors are sensitive to temperature, the more is the thermal FB. LM 3675 is 6 years early version and have input bias current , function of Hfe, has the first curve and the second is 86. So one can estimate that the 75 should have higher thermal FB. No, it is the contrary.

[vk6zgo]

The LM3886 or TDA7293, get the distortion increased with decrease in frequency. See curve bellow. They are opamp type with dominant pole of less than 20hz. The lower is the frequency, the higher is the NFB, so it must continue to decrease.
Thermal feedback?

This is maybe a red herring, but if it is harmonic distortion in a bandwidth limited system, distortion at higher frequencies may appear to be better because the harmonics are "rolled off".

In an earlier existence, I used to travel around AM Broadcast sites doing annual "routine tests".
These were a bit more intensive than those done prior to & after transmission (this was in the days of staffed transmitters in the city & unattended ones in regional centres).

From memory, we would test for THD at 440Hz in the daily tests, & at 60Hz, 440Hz, & 3kHz in the annual ones.
By & large, the THD of the properly engineered Transmitters became worse at the lower & higher test frequencies, although still in spec.

There were a few transmitters which were quite marginal in many other respects & with them, the 60Hz THD was still a bit worse than at 440Hz but the THD at 3kHz was dramatically better.

When we checked the audio frequency response of these "dodgy" transmitters, "the penny dropped". The usual results for the other transmitters in the network was response, essentially "flat" out to 10kHz, but these ones started to fall quite sharply above 6kHz. The response to the 2nd harmonic of 3kHz was thus very much reduced.

[PCB.Wiz]

Thermal feedback?

I'm not sure I'd call it thermal feedback, but certainly any die works as a thermal low pass filter, and lower frequencies never get to an even die temperature.

You can see the same sub-100Hz degrade effect popping up in PSRR numbers on voltage regulators, operating at load currents.
Here is a recent TI regulator

[srb1954]

The more the low level transistors are sensitive to temperature, the more is the thermal FB. LM 3675 is 6 years early version and have input bias current , function of Hfe, has the first curve and the second is 86. So one can estimate that the 75 should have higher thermal FB. No, it is the contrary.

The input bias current is not likely to have a major effect as the test circuit specifies relatively low value input resistors so the offset voltage developed across them is also relatively small and thermal variations of Hfe are less significant.

Vbe matching between the input stage transistors is more likely to be important than Hfe matching. With the recommended non-inverting amplifier configuration, any variation of the input offset voltage of the differential pair input is outside the feedback loop and hence any distortion thus introduced cannot be reduced with feedback.

I don't think you can draw any conclusions by comparing the curves of two different devices as the internal layout may be significantly different causing any thermal feedback issues to also be different.

[srb1954]

This is maybe a red herring, but if it is harmonic distortion in a bandwidth limited system, distortion at higher frequencies may appear to be better because the harmonics are "rolled off".

In this case, the test specifications seem to specify a BW of 80kHz for the distortion measurements so any roll-off of the harmonics would only start to appear for operating frequencies at the very top end of the audio range.

[Kokoriantz]

It.is written on the graph, no filter.

[Kokoriantz]

The TDA curve shows 50w and 100mw. I doubt the thermal FB is the cause.

[Zero999]

The plot is for noise + distortion.

The increase at lower frequencies is mainly due to 1/f noise.

https://www.analog.com/en/resources/analog-dialogue/articles/understanding-and-eliminating-1-f-noise.html

[Kokoriantz]

The 100mw to be higher than 50w is indeed due to noise, but decreasing frequency with constant power, the noise doesn't account. With 100mw 8 ohm, there is no thermal variation neither.I

[Kokoriantz]

The LM3886 and 3875 are identical except the first I suppose, have 2x2 output transistors as it can load 4 ohms.

[Kokoriantz]

Maybe, the internal compensation is acting on high frequencies to decrease the open loop output impedance but less and less at lower. Bellow is tl072

[Zero999]

The 100mw to be higher than 50w is indeed due to noise, but decreasing frequency with constant power, the noise doesn't account.

Yes it does. Did you read the link I posted? The increase in noise at lower frequencies is common to all amplifiers, barring chopper stabalised amplifiers.

With 100mw 8 ohm, there is no thermal variation neither.I

1/f noise isn't hugely temperature dependant.

[Kokoriantz]

The noise character is  independent to operating frequency. It's frequency components remain the same. Or else any opamp would have higher distortion at lower frequency as is THD+N for low level output.

[Zero999]

The noise character is  independent to operating frequency. It's frequency components remain the same. Or else any opamp would have higher distortion at lower frequency as is THD+N for low level output.

No, noise and distortion are both frequency dependant.

The graphs you've posted combine both noise and distortion. They are effectively % error in the output signal vs frequency.

At higher frequencies, the distortion dominates.

At low frequencies, 1/f noise dominates.

[Kokoriantz]

I found measurement THD+N vs frequency. I don't trust much the author, as he is testing a cheap kit and he sells $50 a mono PCB board.

[Kokoriantz]

More reliable source of measurment from JAT youtube shows 2 board his and kit with different gains for 4 and 8 ohm.
https://youtu.be/P_vBeKr8K-c?si=PykBlZIrmJmtWAtC

[whollender]

No, noise and distortion are both frequency dependant.

The graphs you've posted combine both noise and distortion. They are effectively % error in the output signal vs frequency.

At higher frequencies, the distortion dominates.

At low frequencies, 1/f noise dominates.

I initially blamed the 1/f noise as well, but for a THD+N measurement, the measurement bandwidth is not dependent on the signal frequency, so the noise contribution should be constant regardless of the signal frequency even though the noise amplitude is not flat across frequency.

It's possible that there is some kind of second order effect going on, though.

[David Hess]

Just to be clear, thermal feedback increases distortion by lowering open loop gain.

Since the measurement is of total harmonic distortion plus noise, both thermal feedback and flicker noise should be contributing.

Audio power amplifiers usually include resistive emitter degeneration in this first stage to allow more decompensation for a higher slew rate and greater full power bandwidth.  This increases the input stage white noise, decreasing the contribution from flicker noise, but it will still be there.  I guess check the input noise graph, if available, to get an idea of the contribution from flicker noise.  Otherwise you will have to measure it.  A couple of spot noise measurements are enough, but will require some math to work out the white noise level and flicker noise corner frequency.

[Zero999]

What's the thermal time constant of the IC and heat sink? I would have thought it's tens of seconds, possibly even minutes.

Thermal feedback might be an issue at very low frequencies and at high power levels, but the graph of TDA7293, shows the THD+N plots for 100mW out and 50W are essentially the same, indicating thermal feedback is insignificant from 20Hz to 20kHz.


Note the x-axis is in kHz.

[Kokoriantz]

@David Hess, why the open loop decreases by the thermal feedback? BJTs increase the Hfe by temperature.
I am more convinced by the explanation given by @srd1954. The LM3886 has four input transistors, if the thermal wave increases the Hfe of one of them more then the other 3, there is an unbalance. The designer must have privileged the feedback to be negative.
I will harvest the thermal ac component from the mute circuit to inject it in either inputs .