Study: Headphone Amp Gains: Low or High?

[Zilfallion]

What are the low/high gain values?

Cannot help but notice that voltage difference. To my uneducated guess: "Magni 3 gives twice the voltage out while only slightly worse noise than Atom". Can u elaborate?

I can answer both of these with the same answer. The Magni runs 2x low-gain, and 7x high-gain. The Atom runs 1x low gain, and 4.5x high-gain.

 

[DonH56]

For the dB-minded folk:

1x = 0 dBV (0 dBW)​

2x = 6 dBV (3 dBW)​

4.5x = 13 dBV (6.5 dBW)​

7x = 17 dBV (8.5 dBW)​

 

[solderdude]

I wrote something about dB's but the dBW is not in there.
Hmm now I must revisit that page..

 

[Jimster480]

This is what I always have wanted to tell those who believe that much more unused power is good sound quality or something like encouraging headphone's potential ability. I heard some guy who works at pro studio says that they prefer low gain for convenient volume control and good SNR. Today, so many headphones and IEM tend to be sensitive.(I don't think HE6 is ordinary) In my humble experience, many powerful amplifiers need attenuator like ifi's IEM match. 12V output power? I think it's ridiculous.

This is the case with amps like the Magni 3.
But its not the case for the A30 or the JDS O2 or the THX AAA 789.

 

[trl]

Very neat, interesting to see how the noise floor rises between low and high gain.
[...]

It's normal for the background noise to increase with gain, because the noisiest stage from inside an amplifier is usually the input stage and/or the voltage amplification stage (the gain stage). I've seen that happening with my Objective2 headamp and with 1X gain there's absolutely no background noise on my 16 Ohms/100dB SPL IEMs, instead when switching to 3.5X of gain a tiny little bit noise appears on my ears (volume to the max., no input source connected, inputs shorted to ground).

This noise is coming from the internal wires, from volume potentiometer and input resistors (Johnson thermal noise), from the PCB traces and from the active components (opamps and transistors).

Eg.: If an opamp has a 10uV RMS noise on its output, then a gain of 3X will output 30uV RMS noise. Adding 6uV RMS from, let's say, a 100 KOhms potentiometer (for internal temp of 50C, based on http://www.sengpielaudio.com/calculator-noise.htm) means an output noise of 36uV RMS. Every component is adding noise inside a circuit, so everything counts.

Some docs that might help below:
- http://dicks-website.eu/noisecalculator/index.html
- http://www.muzique.com/schem/opcalc.htm
- http://www.ti.com/lit/an/slyt094/slyt094.pdf
- http://www.ti.com/lit/an/slva043b/slva043b.pdf
- https://www.analog.com/media/en/reference-design-documentation/design-notes/dn015f.pdf
- https://www.analog.com/media/en/training-seminars/tutorials/MT-049.pdf

Worth mentioning that insufficient opamp bypassing and a noisy PSU will induce some noise within the opamps and transistors across the board. Also, EMI/RFI might get injected into the circuit too, and some lower harmonics could get into the audible band. This also depends on the CMRR/PSRR of the opamp itself, based on the frequency band.

 

[DataX]

It's normal for the background noise to increase with gain, because the noisiest stage from inside an amplifier is usually the input stage and/or the voltage amplification stage (the gain stage). I've seen that happening with my Objective2 headamp and with 1X gain there's absolutely no background noise on my 16 Ohms/100dB SPL IEMs, instead when switching to 3.5X of gain a tiny little bit noise appears on my ears (volume to the max., no input source connected, inputs shorted to ground).

This noise is coming from the internal wires, from volume potentiometer and input resistors (Johnson thermal noise), from the PCB traces and from the active components (opamps and transistors).

Eg.: If an opamp has a 10uV RMS noise on its output, then a gain of 3X will output 30uV RMS noise. Adding 6uF RMS from, let's say, a 100 KOhms potentiometer (for internal temp of 50C, based on http://www.sengpielaudio.com/calculator-noise.htm) means an output noise of 36uV RMS. Every component is adding noise inside a circuit, so everything counts.

Some docs that might help below:
- http://dicks-website.eu/noisecalculator/index.html
- http://www.muzique.com/schem/opcalc.htm
- http://www.ti.com/lit/an/slyt094/slyt094.pdf
- http://www.ti.com/lit/an/slva043b/slva043b.pdf
- https://www.analog.com/media/en/reference-design-documentation/design-notes/dn015f.pdf
- https://www.analog.com/media/en/training-seminars/tutorials/MT-049.pdf

Worth mentioning that insufficient opamp bypassing and a noisy PSU will induce some noise within the opamps and transistors across the board. Also, EMI/RFI might get injected into the circuit too, and some lower harmonics could get into the audible band. This also depends on the CMRR/PSRR of the opamp itself, based on the frequency band.

Very cool, thank you for those links. This is way beyond my current understanding, but I feel like I'm learning a little bit more every day.

Since we're on the topic, here's a random question I've had on the back of my mind for a bit: If there's a significant amount of thermal noise being introduced, would it make any sense to use sub-ambient cooling in an amplifier circuit, like a Peltier/TEC? Could this make any measurable difference to the noise floor?

Further, what would be the theoretical noise floor limit that can be achieved? I.e. with the most precise analyzer and a simple loopback, what do you end up measuring? Background RFI / EMI noise?

 

[solderdude]

That would consume a lot of power in case you would want to cool a lot of components and gain maybe 1 dB or so in a circuit that most likely has noise levels below the audible limit already.

Maybe mic. and cartridge pre-amps could possibly benefit slightly.
Noise from vinyl and noise from the mic itself (and picked up ambient noise) are most likely the worst offenders though.

Let's just say that there is a good reason there isn't any audio gear using (stacked or not) TECs that I know of.
Also don't know any prof gear doing this.
If it were very beneficial it would be used already.

Keeping parts below -40oC in a home environment would be a neat trick anyway and require a lot of heatsinking.

Ovens for X'tals ... yes.
Some measuring equipment... yes.
Some photo sensitive parts that need to collect just a few photons ... yes.

For RFI/EMI we have screening. Although I do realize good screening isn't that easy.

 

[DonH56]

The largest noise source in audio systems is usually not thermal noise but shot, flicker, switching, etc. noise from the active devices.

Peltier coolers are power-hungry and add quite a bit of noise on (of) their own. We use them in IC testing and they are PITA to clean up -- the controllers pulses the thermal heads and inject noise into the die we are testing. Great.

And when you head below 0 degC you have to deal with moisture/ice buildup. Dry air, add that to your media room build...

Bottom line: Wouldn't help much and not practical.

I've forgotten the theoretical thermal noise floor, seems like around -174 dBm/Hz but I do not recall off-hand what resistance and temperature that was at. Assuming 1 ohm at room temp (300 K) -- pretty sure that is right but plenty here who can catch me if I'm wrong -- then that gives a noise floor around -130 dBm for a 20 kHz bandwidth. (dBm = dB referenced to 1 mW.) I got to play around with superconductors once upon a time and that was (ahem) cool, but it is a huge challenge getting signals to/from the Dewar or whatever you use to cool the things.

 

[dc655321]

I've forgotten the theoretical thermal noise floor, seems like around -174 dBm/Hz but I do not recall off-hand what resistance and temperature that was at. Assuming 1 ohm at room temp (300 K) -- pretty sure that is right but plenty here who can catch me if I'm wrong -- then that gives a noise floor around -130 dBm for a 20 kHz bandwidth. (dBm = dB referenced to 1 mW.)

Looks like your recollection is right on the money

 

[DonH56]

Thanks -- I usually double-check as I have that number and the old 4 nV/rtHz number for a 1 k resistor rattling around my little pea brain, never sure when they'll get scrambled (numbers or brains).

 

[Graph Feppar]

This is why I advocate for low gain,having more gain than necesarry degrades performance.Good thing is most modern amps have gain switch so both high sensitivity and low sensitivity headphone users can be satisfied.

 

[rmo]

This is an analysis of whether you are better off using low or high gain mode in a headphone amplifier in their overlapped region. Traditional rule of electronic design says there is no free lunch: higher gain means higher noise. While this has become the "conventional wisdom" and one that I routinely state myself, forum member asked if there is any hard data to back this. So I decided to test the theory on two headphone amplifiers: the Schiit Magni 3 and JDS Labs Atom.

View attachment 18329​

The test matrix here is infinite in scope. What volume does one choose for each gain to test? After pondering for a second or two , I decided to go the defensible route of setting low gain to max and then matching the same in high gain. Both of these amplifiers have analog volume controls and in high gain, they can be touchy as far as getting accurate levels out of them but I managed to get close enough.

Let's see what the measurements say.

Measurements
For these tests, I chose to use 300 ohm test load as that is in my analyzer and hence, higher fidelity than my external dummy load.

Here is the dashboard view of Schiit Magni 3 in low gain at its maximum value:
View attachment 18330

Now let's switch to high gain while achieving the same 3.84 volt output:

View attachment 18331

We take a 3 dB hit. The impact on SINAD is not as large as one expects because the performance of Magni 3 is distortion limited. If you look at the noise floor in FFT in top right, you can see the large increase in noise floor (about 20 dB).

Note also that channel matching suffered a bit in high gain. Slight inaccuracies in the volume potentiate translates into larger errors in high gain mode although obviously this is situation specific.

In low gain mode, the JDS Labs output is much lower:
View attachment 18332

Performance is so amazing that the noise floor falls off the bottom of the FFT produces superlative SINAD of 115.

Let's switch to high gain and match levels:
View attachment 18333

We take a considerable hit to the tune of 13 to 14 dB. As with Schiit Magni 3, our noise floor rises by good bit (around 10 dB).

So I think we have our answer.

Conclusions
Given a choice, use the low-gain setting of the headphone amplifier unless actual measurements stipulate otherwise. In the case of both Schiit Magni 3 and JDS Labs Atom, this is indeed the wise strategy.

----
If you like these kind of articles, please consider donating funds to support the forum using:
Patreon: https://www.patreon.com/audiosciencereview)
or upgrading your membership here though Paypal (https://audiosciencereview.com/foru...eview-and-measurements.2164/page-3#post-59054).

Without looking at any measurements I have never listened to any headphone amp that sounded better in high gain mode.

 

[Roen]

@amirm

With regards to the THX AAA 789, this particular amp uses a less than unity gain stage. Without regards to the crosstalk issue (unless it is related), is it better to use unity gain, or the lowest possible gain that an amp offers, including gains less than unity?

 

[trl]

I wonder how they achieved this 0.66X gain, but if resistive divider was used then I'll recommend using the unity gain instead of 0.66X.

 

[amirm]

@amirm

With regards to the THX AAA 789, this particular amp uses a less than unity gain stage. Without regards to the crosstalk issue (unless it is related), is it better to use unity gain, or the lowest possible gain that an amp offers, including gains less than unity?

I am not sure without measuring it.

 

[mickeyd123]

So am I better off using the lowest gain or middle gain on the THX AAA 789?

 

[trl]

If you have enough juyce for your cans on the low-gain then you should use it.

Basically, lowest gain that makes volume pot to sing your cans very loud around 2-3 o’clock. This will also protect your hearing (if you’re listening to 9 o’clock and someone else rotates the knob to the max. then it’s not good). Although, channel imbalance is not perfect below 9 o’clock.

 

[mickeyd123]

If you have enough juyce for your cans on the low-gain then you should use it.

Basically, lowest gain that makes volume pot to sing your cans very loud around 2-3 o’clock. This will also protect your hearing (if you’re listening to 9 o’clock and someone else rotates the knob to the max. then it’s not good). Although, channel imbalance is not perfect below 9 o’clock.

Thanks!

 

[Jimster480]

If you have enough juyce for your cans on the low-gain then you should use it.

Basically, lowest gain that makes volume pot to sing your cans very loud around 2-3 o’clock. This will also protect your hearing (if you’re listening to 9 o’clock and someone else rotates the knob to the max. then it’s not good). Although, channel imbalance is not perfect below 9 o’clock.

Basically this.
I use it on the lowest gain at like 1-3 oclock for my planars.

 

[JohnYang1997]

Volume pot at 50% position has the highest noise. 10k pot is better than 100k or 50k pots. But still 2.5k resistance has quite a lot of noise comparing to the noise performance of best opmaps. That's not so much if you run at 25% or lower. Pots used for audio are log/audio taper so 12 oclock is still somewhat low. But i prefer 1k ohm pot. As long as you don't have 1k impedance headphones, 1kohm will not introduce more crosstalk. In this case the system noise will be limited to 0.8uV rms (20-20khz)or lower.