Decoupling/bypass & signal path capacitors for op-amps (NE5532 in Aiyima A07 vs. TPA3255evm)

[Cbdb2]

Page 5 shows how a wima changes impedance with freq. and size. The larger caps have a higher Q (more inductance), so there effectiveness varies more with freq. The larger caps also couple to the PCB more.

https://www.mouser.ca/datasheet/2/440/e_WIMA_MKP_4-1139861.pdf

So just adding large caps is throwing the dice. I find it hard to believe you always get better sound.

 

[Cbdb2]

The TPA3255evm was designed by Texas Instruments. It uses NE5532 in the input stage. The designers thought it would be a good idea to add 10 μF to the 0.1 μF. So why does TI think the additional 10 μF is necessary? Does it improve the performance?

Its a class D Amp so those caps were probably chosen to work at the Amps switching freq and harmonics. That does not mean its a universal solution. The tech note from TI I posted above explains this.

 

[Zapper]

So you just wilynily add caps?

Also from:https://www.ti.com/lit/an/sloa069/sloa069.pdf

Misconception number 2: Most designers think that when poor decoupling is suspected,
increasing the value of capacitance will produce more attenuation. The exact opposite is true.
When poor decoupling of high frequencies is suspected, decrease the value of capacitance (or
preferably, do the homework and look up the series self-resonance)

But next sentence is also relevant:

If more than one frequency (or band of frequencies) is present in the system, two or more capacitors may be required to properly bypass. Each capacitor is then targeted at a different frequency range.

 

[Dimitri]

So why does TI think the additional 10 μF is necessary?

Here is their story and they are sticking to it: ( https://www.ti.com/lit/an/sloa069/sloa069.pdf )
"
Texas Instruments includes electrolytic capacitors on evaluation boards because there is a chance they will be
operated off a switching power supply, or operated near AM stations with long power-supply leads.
Most laboratory applications, however, are well-shielded from AM radio stations, and most laboratory
power supplies are not switching supplies. Therefore, electrolytic capacitors are not usually required"

I liberally highlighted the fun parts

 

[Cbdb2]

But next sentence is also relevant:

Yes, if the offending freqs. are far apart, you need more than one filter. The key is knowing what freqs. are the problem and using the proper caps for them. Just adding capacitors is not the way to do it. AM radio and switching supplies are lower freqs. If the problem is cell phones electrolytics are useless.

 

[LTig]

The .1u on the power pin is not for instability its to filter RF out of the DC, the PSRR becomes low at these high freqs.
From: https://www.ti.com/lit/an/sloa069/sloa069.pdf

On the other hand I have seen opamps (TL074 AFAIR) oscillating without those caps between power supply pin and ground, and adding them calmed them down.

 

[BR52]

The pc- board design is also a key component (ground plane, multi layer, and so on )play a important rule in case of oscillations, RF immunity, voltage spikes. Decoupling can be more than only place a capacitor especially in switching circuits.

 

[Zapper]

The TPA3255evm was designed by Texas Instruments. It uses NE5532 in the input stage. The designers thought it would be a good idea to add 10 μF to the 0.1 μF. So why does TI think the additional 10 μF is necessary? Does it improve the performance?

It looks like the TI designer of the TPA3255evm was concerned about the high power switching of the TPA3255 coupling into the low signal NE5532. Significant bill of materials is used for the +12V-OA opamp suppy. First a LM5010ASD buck steps down the input voltage to 15V, then a LM2940IMP-12 LDO is used to generate a cleaner 12V supply, which is further filtered by an LC filter, which includes 10uF and 0.1uF caps at each op amp.

Whether the 10uF improves performance is impossible to determine by looking at schematics and PC board layouts, especially since we have neither for the A07. The need for the capacitor would depend a lot on PC board layout, which differs between the Aiyima and TI designs. One would have to put it on an audio analyzer and try both configurations to see if the 10uF made a measurable difference.

 

[Zapper]

The .1u on the power pin is not for instability its to filter RF out of the DC, the PSRR becomes low at these high freqs.
From: https://www.ti.com/lit/an/sloa069/sloa069.pdf

I think you are misapplying this app note. It is specifically for "high speed op amps" as stated in the title. TI defines those as having "gain bandwidth product (GBW) ranging from 50 MHz to 8 GHz". The author refers to "microwave capacitors" in multiple places. This is not the type of op amp we are discussing here. The GBW of the NE5532 is 10MHz, and there are no microwave capacitors on either the TPA3255evm nor the Aiyima board.

The decoupling capacitor is very much for stability. The GBW of the op amp is 10MHz. The self resonance frequency of the C0603C104K5RACTU cap is 20-30MHz, so it is not providing in-band filtering at its self resonance. It is located close to the op amp to minimize the PC board inductance in series with the op amp supply, to avoid an unintentional feedback path in the op amp due to inductive voltage drop on the supply.

When the op amp drives its output load, the current drawn from the supply will generate a voltage across any series inductance in the supply lead. The voltage across the inductor leads the current, but as the voltage drop is inverted with respect to the output voltage, that creates a lag instead of a lead relative to the output voltage. This voltage couples into the internal signal path, especially at high frequencies where the PSRR is poor. The resulting phase of the injected signal now depends on the output load impedance, the impedance at the supply pin, and the internal circuit topology of the op amp. Two of these things are out of the designers hands. The effect can be highly variable, but it will likely reduce phase or gain margin. Some op amps can definitely be observed to oscillate under such conditions.

This is an uncontrolled nested feedback problem that no IC designer wants to deal with. Requiring a 0.1uF ceramic cap very close to the supply pins bounds this problem and largely eliminates PCB inductance concerns at the supply pin.

On the other hand I have seen opamps (TL074 AFAIR) oscillating without those caps between power supply pin and ground, and adding them calmed them down.

Agreed, those caps are primarily for stability.

 

[Zapper]

The TI TPA3255EVM uses 10uF electrolytics in many places in the signal path. I'm not a capacitor expert at all, but I thought this was not considered good practice*.

* P.S. As usual, TI has an app note on that. Electrolytics are worse than film (but not as bad as I thought). Still, definitely not audiophile quality.

 

[LTig]

The TI TPA3255EVM uses 10uF electrolytics in many places in the signal path. I'm not a capacitor expert at all, but I thought this was not considered good practice*.

This is an urban myth WRT coupling caps. Douglas Self has layed it to rest. Distortion of elcaps rises with the voltage over it. To reduce it make the elcap large enough so that the lowest frequency does not create more than a few mV over it.

For filtering though you cannot do this and hence elcaps are a no go.

 

[Zapper]

This is an urban myth WRT coupling caps. Douglas Self has layed it to rest. Distortion of elcaps rises with the voltage over it. To reduce it make the elcap large enough so that the lowest frequency does not create more than a few mV over it.

For filtering though you cannot do this and hence elcaps are a no go.

That's a good point, and it is discussed in the TI app note I linked. The plot I copied was done at with 1uF @ 1Vrms with 2.5k series resistance. The TPA3255EVM circuit has a 10uF & 10k series resistance, so for a given frequency @ 1V input, the voltage on the cap is 1/40 compared to the plotted data - about 80mV @ 20Hz. So the distortion will be much less than that plot.

However, the disortion of el cap is still higher than film at all freqs, so the designer looking for the best ASR panther shouldn't use the electrolytics.

 

[LTig]

That's a good point, and it is discussed in the TI app note I linked. The plot I copied was done at with 1uF @ 1Vrms with 2.5k series resistance. The TPA3255EVM circuit has a 10uF & 10k series resistance, so for a given frequency @ 1V input, the voltage on the cap is 1/40 compared to the plotted data - about 80mV @ 20Hz. So the distortion will be much less than that plot.

However, the disortion of el cap is still higher than film at all freqs, so the designer looking for the best ASR panther shouldn't use the electrolytics.

Douglas Self measured THD of less than 0.0003% which was in fact the limit of his AP-2702, hence it could be even lower. That was with different elcaps (47 to 1000 μF) for voltages at 80 mV rms over the cap (10 V rms with a load of 1 kΩ), and the latter reached this down to 20 Hz. See Small Signal Audio Design, 2010, Figure 2.17 on page 60.

 

[Cbdb2]

I think you are misapplying this app note. It is specifically for "high speed op amps" as stated in the title. TI defines those as having "gain bandwidth product (GBW) ranging from 50 MHz to 8 GHz". The author refers to "microwave capacitors" in multiple places. This is not the type of op amp we are discussing here. The GBW of the NE5532 is 10MHz, and there are no microwave capacitors on either the TPA3255evm nor the Aiyima board.

The decoupling capacitor is very much for stability. The GBW of the op amp is 10MHz. The self resonance frequency of the C0603C104K5RACTU cap is 20-30MHz, so it is not providing in-band filtering at its self resonance. It is located close to the op amp to minimize the PC board inductance in series with the op amp supply, to avoid an unintentional feedback path in the op amp due to inductive voltage drop on the supply.

When the op amp drives its output load, the current drawn from the supply will generate a voltage across any series inductance in the supply lead. The voltage across the inductor leads the current, but as the voltage drop is inverted with respect to the output voltage, that creates a lag instead of a lead relative to the output voltage. This voltage couples into the internal signal path, especially at high frequencies where the PSRR is poor. The resulting phase of the injected signal now depends on the output load impedance, the impedance at the supply pin, and the internal circuit topology of the op amp. Two of these things are out of the designers hands. The effect can be highly variable, but it will likely reduce phase or gain margin. Some op amps can definitely be observed to oscillate under such conditions.

This is an uncontrolled nested feedback problem that no IC designer wants to deal with. Requiring a 0.1uF ceramic cap very close to the supply pins bounds this problem and largely eliminates PCB inductance concerns at the supply pin.


Agreed, those caps are primarily for stability.

High speed amp circuits are more susceptible to instability than slow ones so that app note still applies.
The bypass caps shunt high freqs to ground, this increases PSRR in all opamps and can increase stability in some circuits so I would not say primarily for stability, especially in slow audio circuits.

https://www.analog.com/media/en/training-seminars/tutorials/MT-101.pdf

Proper Design Techniques Solve High-Speed Op-Amp Stability Problems

Successful high-speed circuit designs need the right combination of grounding, power-supply bypassing, and decoupling.

www.electronicdesign.com

 

[Zapper]

The bypass caps shunt high freqs to ground, this increases PSRR in all opamps and can increase stability in some circuits so I would not say primarily for stability, especially in slow audio circuits.

No doubt the decoupling capacitor is important for PSRR, noise, output slew rate, and other things.

 

[Zapper]

That was with different elcaps (47 to 1000 μF) for voltages at 80 mV rms over the cap (10 V rms with a load of 1 kΩ), and the latter reached this down to 20 Hz.

It was only the 1000uF 25V capacitor that reached that low level. The smaller values showed distortion at low frequencies.

The 100uF/1kOhm case in Self's data is comparable to the 10uF/10kOhm circuit in the TPA3255EVM, as both will result in the same voltage drop across the capacitor. Self reports 0.0017% distortion at 20Hz.

So my opinion is only reinforced by this data. Electrolytic caps in the audio signal path are not good practice (unless they are friggin huge).

 

[retroflex]

It was only the 1000uF 25V capacitor that reached that low level. The smaller values showed distortion at low frequencies.

The 100uF/1kOhm case in Self's data is comparable to the 10uF/10kOhm circuit in the TPA3255EVM, as both will result in the same voltage drop across the capacitor. Self reports 0.0017% distortion at 20Hz.

So my opinion is only reinforced by this data. Electrolytic caps in the audio signal path are not good practice (unless they are friggin huge).

A 1000uF 25V capacitor is only like ø10x20mm. The cheapest 100uF polypropylene capacitor on Mouser is significantly bigger and 10 times the price, for no measurable improvement. That's money you could be spending on other components. And to put the whole thing in perspective, @StigErik measured >0.2% THD at 20 Hz with his 24 (!) subwoofer setup, 100 times more distortion than the 10uF electrolytic capacitor of the TPA3255EVM can be expected to cause. So in conclusion, it's easy to achieve measurable perfection with electrolytic dc-blocking capacitors, and even if you don't, they are unlikely to cause any audible problems even in the most extreme systems.

 

[Zapper]

it's easy to achieve measurable perfection with electrolytic dc-blocking capacitors, and even if you don't, they are unlikely to cause any audible problems even in the most extreme systems.

Oh I agree with that. This is all measurbating - it's all far below audible distortion.

 

[DonH56]

The impact of electrolytic capacitors in the signal path depends upon their biasing and the impedances on either side (which also determine the corner frequency). I don't think it is accurate to make a blanket statement that smaller than so-and-so produces too high distortion.

 

[007Shortz]

Are capacitors in the signal path needed for balanced/symmetrical/differential input stages?