I tried playing with R1. Reducing it to 1K drops the output offset to ~2mV. Dunno what it does over temp, but so far so good.
I did not look at the +IN & -IN resistances. It did not occur to me that you would have unbalanced these.
You had 1k8||200r = 180r on -IN and 47k2 on +IN.
Now you have 1k8 on -IN and 1k2 on +IN.
No wonder you had output offset.
The closer you can operate each Long Tail Pair in balance the lower becomes the tempco.
Why have you chosen to operate with mixed AC and DC coupling?
You had 1k8||200r = 180r on -IN and 47k2 on +IN.
Now you have 1k8 on -IN and 1k2 on +IN.
No wonder you had output offset.
The closer you can operate each Long Tail Pair in balance the lower becomes the tempco.
Why have you chosen to operate with mixed AC and DC coupling?
Last edited:
My thought was that I could dc couple the feedback loop on account of the gain setting being reasonably low. It lets me get rid of a large and expensive cap, as long as I tolerate higher DC offset. I couldn’t make that argument for the input though, as the source may have its own output cap.
So once I decided I needed the input cap, the compromise is between noise performance (low resistance), input cap size and pass band frequency. Unbalancing Rin and Rf allows me to trade offset for cap size and frequency for offset.
I think the easy way is just to dc couple the input.
So once I decided I needed the input cap, the compromise is between noise performance (low resistance), input cap size and pass band frequency. Unbalancing Rin and Rf allows me to trade offset for cap size and frequency for offset.
I think the easy way is just to dc couple the input.
Hi Suzy,
err, not a good plan. I have seen preamps in for service that were putting out DC. One memorable one (Krell system) burned out a PSB Stratus Gold (version II) woofer and crossover.
The amount of noise you're talking about will not be something you would notice. What size of coupling cap were you considering? The 10uF plastic (Wima or other) seems to be pretty darned good. If you can afford more space for this you can get even higher values.
-Chris
err, not a good plan. I have seen preamps in for service that were putting out DC. One memorable one (Krell system) burned out a PSB Stratus Gold (version II) woofer and crossover.
The amount of noise you're talking about will not be something you would notice. What size of coupling cap were you considering? The 10uF plastic (Wima or other) seems to be pretty darned good. If you can afford more space for this you can get even higher values.
-Chris
add a DC blocking capacitor to the feedback lower leg.
A reformed electrolytic is OK in this position. You can add on a pair of inverse parallel signal diodes to prevent damage to the electro during any abusive incident.
Once that is added, you can then match up in +IN and -IN resistances.
The alternative is to go all DC coupled and use offset detection to trigger protection, along with a DC servo to control the offset.
I’ve left enough space on the board for a 1uF Wima MKS2. Granted, a board respin is trivial, and likely to be done anyway to add more output drivers, for example, so there’s scope to provide room for a larger input coupling cap and an electrolytic to the feedback side.
Indeed I really should do this, as I haven’t the slightest clue what effect an electrolytic in the feedback loop will have on distortion. I’m guessing, like everything, it can be made suitably small by using a large enough cap.
Electrolytic are cheap. Alas big film caps aren’t. 10uF MKS2 caps run ~$4 each, which while not huge is ten times the cost of any other individual part.
I’m suspecting the most cost effective solution to minimising offset while maintaining noise performance is simply throwing more transistors at the problem and doing a DC servo.
Indeed I really should do this, as I haven’t the slightest clue what effect an electrolytic in the feedback loop will have on distortion. I’m guessing, like everything, it can be made suitably small by using a large enough cap.
Electrolytic are cheap. Alas big film caps aren’t. 10uF MKS2 caps run ~$4 each, which while not huge is ten times the cost of any other individual part.
I’m suspecting the most cost effective solution to minimising offset while maintaining noise performance is simply throwing more transistors at the problem and doing a DC servo.
Hi Suzy,
That would work as long as you didn't run into common mode problems.
I would use the $4 capacitor and be done with it. No issues with a DC offset voltage on your input and using the film type capacitor is better than an electrolytic in most cases. I'd hate to be guessing about this while troubleshooting an odd problem.
I am seeing more preamps without an output capacitor these days. They generally use a servo so the DC output is on the order of mV. This has some validity since you don't know what load impedance you're going to see. The output capacitor would have to be very much larger than needed with a higher impedance input (which you can control). So your only known factor is the input impedance of your amplifier. This would then be the most logical place to put in a DC blocking capacitor.
-Chris
That would work as long as you didn't run into common mode problems.
I would use the $4 capacitor and be done with it. No issues with a DC offset voltage on your input and using the film type capacitor is better than an electrolytic in most cases. I'd hate to be guessing about this while troubleshooting an odd problem.
I am seeing more preamps without an output capacitor these days. They generally use a servo so the DC output is on the order of mV. This has some validity since you don't know what load impedance you're going to see. The output capacitor would have to be very much larger than needed with a higher impedance input (which you can control). So your only known factor is the input impedance of your amplifier. This would then be the most logical place to put in a DC blocking capacitor.
-Chris
D.Self has given a lot of information on this topic starting back in the mid 1990s. That topic also comes up in his power amp books and in his filter book.
The DC blocking capacitor in the NFB lower leg has virtually no DC across it.
If it is sized correctly to pass the audio signal, then you also find it has virtually no audio signal across it.
With zero voltages across it cannot generate distortion to add to the signal injected into the -IN node.
To pass the audio signal without distortion the DC blocking capacitor should have an F-3dB frequency approximately one decade below the audio frequency band, when a low disortion capacitor is used.
If you use an electrolytic, (higher distortion if there is an AC voltage across it) then move the F-3dB frequency down another octave.
eg for a MKS/MKT/MKP blocker set the F-3dB to 2Hz to pass a >=20Hz audio signal.
For an electrolytic set the F-3dB frequency to 1Hz to pass a >=20Hz audio signal.
To minimise the AC signal across this capacitor the input filter should be set to ~ 3Hz
When you use an electrolytic you must reform it before assembly. Otherwise it will leak for days/weeks/months while it reforms naturally with the very low DC voltage it sees in circuit. And it must be oriented to match the small DC voltage that exists. Or use a Bi-Polar which is bigger and more expensive.
Last edited:
Does this make sense?
"If we want to listen at 90 dB, peaks will reach about 110 dB. At this playback level, the distortion must be lower than -110 dB (0.0003%) to absolutely guarantee that it is inaudible."
Interpreting THD Measurements - Think dB not Percent! - Benchmark Media Systems, Inc.
//
"If we want to listen at 90 dB, peaks will reach about 110 dB. At this playback level, the distortion must be lower than -110 dB (0.0003%) to absolutely guarantee that it is inaudible."
Interpreting THD Measurements - Think dB not Percent! - Benchmark Media Systems, Inc.
//
Last edited:
I belive it has to do with the spectral divison of the overtones. Higher, odd ones is nasty - and detectable at really low levels.
//
Oh absolutely. But that’s not really the point. An amp that does 0.01% and 50nv/rthz is totally fine, but boring as a design exercise. As several have rightly pointed out, by any reasonable measure I should just buy an lm1875 and go with that.
The enjoyable part for me is pushing the boundaries, for no other reason than I can. If I can measure something I’ll probably obsess over it. My ability to measure small things is getting pretty good these days, and that gives me lots of enjoyment.
The cost requirement is purely arbitrary, but is a really good way for me to keep things from getting out of hand, and keep my relationship happy. It’s why I’m working with 4” drivers, and why I’m going away from exotic lateral mosfets.
- my question was more directed to Markw4.In some cases tones below the threshold of hearing (a statistical value that probably applies to most people, buy maybe not all) can become audible in the presence of other waveforms.
In addition, some or maybe only a few people happen to be good at detecting or noticing very small distortions. It isn't a function of the ears so much as it is how brains happen to be wired neurologically. Some people have perfect pitch, some people notice distortion more so than average. How much more so? Unfortunately, we don't have good research on that. But, we do know that some people can hear undithered 16-bit audio. That's getting down pretty close to -100dB below FS. And, that's probably not the limit.
Some people have very good ability to hear distortion. Most people are in the -21dB category. Here’s a nice tear by Klippel. Use good headphones and a quiet room at night.
Listening Test
I don't think it's the level of harmonic distortion we hear that is maybe unpleasant. I've made big improvements to equipment that reduced THD by a small amount. The difference in sound quality was larger than the numbers would suggest. What I do see changing are the residuals from the THD meter (HP 339A in my case) as seen on a spectrum analyser. A chip amp can sound completely different to a discrete amplifier or a tube amplifier and have very similar THD numbers. Suzy, your 3585B should show this easily if you hang it off a distortion analyser.
-Chris
-Chris