With competently designed audio equipment the minimum THD+N occurs just below clipping. That's true for the Q1MKII, the 10B and all of the Bryston *B amplifiers.
The difference between them is that the 10B has the equal greatest dynamic range of the three and therefore is the most tolerant of sub-optimal signal level, so you've got ±6dB or so to play with when optimising the signal gain through the 10B.
In a professional audio set up the only time I would ever run an amplifier without a 20dB pad in front of it is for a loud live music concert, when the room SPL has to be able to achieve over 100dBspl many meters away from the speakers, and the noise in the speakers won't be heard above the crowd ambiance.
The difference between them is that the 10B has the equal greatest dynamic range of the three and therefore is the most tolerant of sub-optimal signal level, so you've got ±6dB or so to play with when optimising the signal gain through the 10B.
In a professional audio set up the only time I would ever run an amplifier without a 20dB pad in front of it is for a loud live music concert, when the room SPL has to be able to achieve over 100dBspl many meters away from the speakers, and the noise in the speakers won't be heard above the crowd ambiance.
Last edited:
I just found this description of noise in audio devices on Rane's website, which I thought was relevant to understanding the different devices you have:
High-Quality Low-Voltage Audio
Noise
Low noise and low voltage don't like each other. Low voltage usually means portable, and portable always means low current to prolong battery life. You can design low noise and low voltage if you can be a current pig, but if you must have low noise, low voltage and low current -- well, that's difficult.
Everything works against you. The easiest way to make a really low noise op amp is to run as much current as possible through the front-end differential-pair until the silicon glows.
As unintuitive as it may be, a plain resistor, hooked up to nothing, generates noise and the larger the value the greater the noise. It is called thermal noise or Johnson noise (John Bertrand Johnson first observed thermal noise while at Bell Labs in 1927, publishing his findings as "Thermal agitation of electricity in conductors," Phys. Rev., vol. 32, pp. 97-109, 1928), and results from the motion of electron charge of the atoms making up the resistor. All that moving about is called thermal agitation (caused by heat -- the hotter the resistor, the noisier). Therefore quiet designs should use small resistor values, but, alas, small resistor values draw large current, and there goes the battery life. Compromise must ensue. It is difficult to find the perfect balance between small resistor values for low noise and large resistor values for low current consumption. To make it even harder, with most analog circuits small resistor values mean correspondingly large capacitor values. Large capacitor values do not hurt the noise performance but they are physically large and cost more, so you must make a compromise between noise, space and cost (analog design is like that).
Low noise and low voltage don't like each other. Low voltage usually means portable, and portable always means low current to prolong battery life. You can design low noise and low voltage if you can be a current pig, but if you must have low noise, low voltage and low current -- well, that's difficult.
Everything works against you. The easiest way to make a really low noise op amp is to run as much current as possible through the front-end differential-pair until the silicon glows.
As unintuitive as it may be, a plain resistor, hooked up to nothing, generates noise and the larger the value the greater the noise. It is called thermal noise or Johnson noise (John Bertrand Johnson first observed thermal noise while at Bell Labs in 1927, publishing his findings as "Thermal agitation of electricity in conductors," Phys. Rev., vol. 32, pp. 97-109, 1928), and results from the motion of electron charge of the atoms making up the resistor. All that moving about is called thermal agitation (caused by heat -- the hotter the resistor, the noisier). Therefore quiet designs should use small resistor values, but, alas, small resistor values draw large current, and there goes the battery life. Compromise must ensue. It is difficult to find the perfect balance between small resistor values for low noise and large resistor values for low current consumption. To make it even harder, with most analog circuits small resistor values mean correspondingly large capacitor values. Large capacitor values do not hurt the noise performance but they are physically large and cost more, so you must make a compromise between noise, space and cost (analog design is like that).
The choice of resistor values then becomes the deciding factor in selecting the right op amp for each application. Look at the resistor values; if they are very small (like in a mic preamp) then the noise contributed by the op amp becomes critical. However, if the application is active filters, say, and the resistors surrounding the op amp are at least 10 k ohm, then the dominate noise factor becomes their thermal noise, not the op amp's noise. Understanding this simple fact allows you to use low-cost op amps for most of your needs.
Ultimately the performance gets down to how much voltage is available and how low is the noise floor: power supply and noise -- the big two in designing quality audio for IAs (portable audio).
Ultimately the performance gets down to how much voltage is available and how low is the noise floor: power supply and noise -- the big two in designing quality audio for IAs (portable audio).
High-Quality Low-Voltage Audio
This article is misleading. At the intro voltage noise of bjt-inputs is considered. Which indeed can be minimized by max emitter current. But at the end we find a >10kOhm resistor will be the major contributor on noise - which is definitely wrong in the bjt-application mentioned on entry - but right in case of some low-noise FET oder MOSFET input stage - were current noise is low. A strict differentiation between current and voltage noise is mandatory but missing. So it is just another half-truth found on the internet.
Last edited:
So the Fiio's SE output is 6 dBu and the BAL output is 10.6 dBu. Because the 10B has +/- 24V voltage rails for its op amps, those will clip close to that. That's 26.8 dBu. So if I wanted to use the SE output to drive the 10B close to clipping, I would need roughly +20.8 dB gain.
The question is how close to clipping I should be setting it. 1 dB less? 0.8 dB less? 2 dB? 3dB? 0.1 dB? I have no idea what the right value would be here.
OK, so using SE output I want 20 dB gain between the DAC and the 10B.
Converting my listening levels from the BAL output to the SE output, I normally listen at -18 dB, but sometimes at -3 dB, +3 dB, or even +6 dB.
So this means to me that the pad after the 10B and going to the amps should be capable of the following levels:
-38 dB
-23 dB
-17 dB
-14 dB
But I still don't know how hard I should be driving the 10B exactly.
Converting my listening levels from the BAL output to the SE output, I normally listen at -18 dB, but sometimes at -3 dB, +3 dB, or even +6 dB.
So this means to me that the pad after the 10B and going to the amps should be capable of the following levels:
-38 dB
-23 dB
-17 dB
-14 dB
But I still don't know how hard I should be driving the 10B exactly.
With the data available in the specifications for the 10B and the Fiio it is impossible to answer that question.
If you run 20dB of attenuation in front of the amplifier, you can be fairly certain the the noise limit will be somewhere else in your signal chain, i.e. cables, the DSP (which has less dynamic range) or the program material. There is no need for exactness.
On connecting, without a line amp balanced interface, your best connection is probably the bridged headphone output wired correctly, which is (for each channel) as per the diagram #7 here:
RANE Commercial - Knowledge Base
If you want a better understanding of balanced connections there is a good description here:
Balanced Interfaces
Hi cheater,
Normally I wouldn't mention this -- it would be a substantial undertaking, likely fraught with oopsie-doodles -- but you did seem singularly obsessed with the 10B's noise level:
If those images on post 20 are of your unit (or representative of it), there may be modest gains to be had by *de-substituting* transistors. Many of the fitted p/n's say 2N5550 and 2N5400, while the silkscreen reads '5210 and '5087, which are noted low-noise performers. My guess is the '5550 and '5400's were available in narrow ranges of hFE and Vbe, so as to not require the usual careful matching. They are much higher voltage parts, and not specified or sold as low noise units.
The caveats are several (and probably more I haven't thought of):
- top shelf soldering skills are essential
- some Vbe matching of the replacements will probably be required,
and you may have to order excess to achieve it
- careful attention to detail and record keeping also essential
- the result may still not satisfy
If you're determined to attempt this, I recommend locating the section that has the intermittent, variable noise level and start with it. Decide on a weighting (A, C, none) curve and stick with it; don't use any other. Take careful measurements, trying to catch it both malfunctioning and not, and write it all down.
Remove one transistor at a time, fold a piece of masking tape or self-adhesive label over it and mark its component designation (Q301, etc). That way, if your soldering is good enough, you can restore the original state. Determine which ones function in pairs and measure Vbe, and if you can hFE at ~15V. That will give an idea how closely-matched the replacements need to be.
Then be sure to leave a few minutes at the end of each work interval to kick yourself for such a massive undertaking.
Honestly, I would just put a 20dBV pad on every amp channel and not look back. But if you DO undertake this, bet there'll be at least a few of us here that'll send kudos -- which is probably good since it'd be hard to put on your resume.
Normally I'm one of the very last members to recommend wholesale replacements; there was just something about the combination of it being a Bryston piece, the silkscreen printing, and your persistent determination ..
And keep the original transistors organized and available. Unforeseen problems may still pop up.
Cheers
edit: Sorry if I'm out of line here -- it's taken me since post #88 to cobble this up.
Normally I wouldn't mention this -- it would be a substantial undertaking, likely fraught with oopsie-doodles -- but you did seem singularly obsessed with the 10B's noise level:
If those images on post 20 are of your unit (or representative of it), there may be modest gains to be had by *de-substituting* transistors. Many of the fitted p/n's say 2N5550 and 2N5400, while the silkscreen reads '5210 and '5087, which are noted low-noise performers. My guess is the '5550 and '5400's were available in narrow ranges of hFE and Vbe, so as to not require the usual careful matching. They are much higher voltage parts, and not specified or sold as low noise units.
The caveats are several (and probably more I haven't thought of):
- top shelf soldering skills are essential
- some Vbe matching of the replacements will probably be required,
and you may have to order excess to achieve it
- careful attention to detail and record keeping also essential
- the result may still not satisfy
If you're determined to attempt this, I recommend locating the section that has the intermittent, variable noise level and start with it. Decide on a weighting (A, C, none) curve and stick with it; don't use any other. Take careful measurements, trying to catch it both malfunctioning and not, and write it all down.
Remove one transistor at a time, fold a piece of masking tape or self-adhesive label over it and mark its component designation (Q301, etc). That way, if your soldering is good enough, you can restore the original state. Determine which ones function in pairs and measure Vbe, and if you can hFE at ~15V. That will give an idea how closely-matched the replacements need to be.
Then be sure to leave a few minutes at the end of each work interval to kick yourself for such a massive undertaking.
Honestly, I would just put a 20dBV pad on every amp channel and not look back. But if you DO undertake this, bet there'll be at least a few of us here that'll send kudos -- which is probably good since it'd be hard to put on your resume.Normally I'm one of the very last members to recommend wholesale replacements; there was just something about the combination of it being a Bryston piece, the silkscreen printing, and your persistent determination ..
And keep the original transistors organized and available. Unforeseen problems may still pop up.
Cheers
edit: Sorry if I'm out of line here -- it's taken me since post #88 to cobble this up.
Last edited:
So what's the best way to find out the perfect level to drive the 10B at? What I think could possibly be a way: output -1 dBFS sine wave out of DAC -> preamp -> 10B -> (LF/MF/HF) output -> -20 dB pad -> notch filter to get rid of sinewave -> audio line input on my motherboard -> measure the THD products in FFT
BTW there's no DSP involved. It's straight up DAC to the analog crossover. Unless you mean the DSP inside the DAC chip itself.
BTW there's no DSP involved. It's straight up DAC to the analog crossover. Unless you mean the DSP inside the DAC chip itself.
Thanks, yeah, I have a cable like that already. It was store bought and I had to open up the XLR connectors and snip off pin 1 because it was creating a ground loop, but now it's good.
Hi cheater,
Normally I wouldn't mention this -- it would be a substantial undertaking, likely fraught with oopsie-doodles -- but you did seem singularly obsessed with the 10B's noise level:
If those images on post 20 are of your unit (or representative of it), there may be modest gains to be had by *de-substituting* transistors. Many of the fitted p/n's say 2N5550 and 2N5400, while the silkscreen reads '5210 and '5087, which are noted low-noise performers. My guess is the '5550 and '5400's were available in narrow ranges of hFE and Vbe, so as to not require the usual careful matching.
The caveats are several (and probably more I haven't thought of):
- top shelf soldering skills are essential
- some Vbe matching of the replacements will probably be required,
and you may have to order excess to achieve it
- careful attention to detail and record keeping also essential
- the result may still not satisfy
If you're determined to attempt this, I recommend locating the section that has the intermittent, variable noise level and start with it. Decide on a weighting (A, C, none) curve and stick with it; don't use any other. Take careful measurements, trying to catch it both malfunctioning and not, and write it all down.
Remove one transistor at a time, fold a piece of masking tape or self-adhesive label over it and mark its component designation (Q301, etc). That way, if your soldering is good enough, you can restore the original state. Determine which ones function in pairs and measure Vbe, and if you can hFE at ~15V. That will give an idea how closely-matched the replacements need to be.
Then be sure to leave a few minutes at the end of each work interval to kick yourself for such a massive undertaking.
Normally I'm one of the very last members to recommend wholesale replacements; just something about the combination of it being a Bryston piece, the silkscreen printing, and your persistent determination ..
And keep the original transistors organized and available. Unforeseen problems may still pop up.
Cheers
That sounds like fun. At that rate, might be better and easier to clone the board and lay it out with my own transistors. After all that's all there really is on the boards: a bunch of transistors (pretty pretty pretty cheap), a few ceramic capacitors, and some styrene capacitors (probably the only expense, but still cheap in the long run). As well as some resistors and a rotary switch that is not being used (it's always in the same setting). I haven't fully seen this yet but according to the service manual those boards have card edge connectors, so I should be able to just unplug the old one and plug the new one back in. If you look very closely on the first picture on the left you can see the connector and you can see the card edge contacts showing through the pcb material. I don't have a transistor curve tracer yet, so that might be a fun reason to get one.
Maybe while at it, it might be possible to use a different design for the discrete op amps that has increased noise immunity.
Either way it beats messing up the original boards.
Here's the thing though, however. First I should try muting the op amps going up the signal chain to see how that affects the noise output. I would start with the IC op amps, then whatever that SMD board is, then I could go up the chain on the discrete op amps. If anything jumps out at me, then work on that.
What's the best way to do that? Someone on a different list mentioned taking a 100 ohm resistor and shorting each op amp to ground to see what contributes the most noise (if anything), but I don't know if I should be doing that to the output of the op amp, or the input, and if so whether I should be shorting the inverting or non-inverting input.
Last edited:
Now you want to use THD measurements to set levels!?
(Responding to post 94 ..)
(post 95: )
Please forget what you think you know about 'noise immunity' -- it doesn't apply to this problem.
The op-amp sections are all independent AFTER the input buffer. They're not *summing* such that a particular one might contribute more noise than the others.
Regards
(Responding to post 94 ..)
(post 95: )
Please forget what you think you know about 'noise immunity' -- it doesn't apply to this problem.
The op-amp sections are all independent AFTER the input buffer. They're not *summing* such that a particular one might contribute more noise than the others.
Regards
Last edited:
Now you want to use THD measurements to set levels!?
(Responding to post 94 ..)
Following what John said, there is an optimal level at which to drive the input which will produce the least THD+N.
That is not true. Look at the schematic. It shows a single crossover with a low and high pass. Each of those filters has three op amps in series. Additionally, to get three outputs, the crossover doubles up the LP/HP combo. The first LP/HP xo outputs the LP to the LF output on the back, and the HP to the second xo. The second xo outputs to the MF and HF output on the back. So the HF output has been through two high pass filters. That's six op amps in series.
The schematic only shows one of the crossovers because the other one is identical.
The 10B has significantly more dynamic range than your source, which means that there is a wide window of signal level for the 10B where the source will dictate the system THD+N, and that gives you quite a lot of latitude.
I'll have a think about the best gain structure between the Bryston *B amps and the 10B tomorrow, because that is where the ultimate S-N numbers are determined. Then you can look backwards to see what the gain margin is for consumer audio toys.
I'll have a think about the best gain structure between the Bryston *B amps and the 10B tomorrow, because that is where the ultimate S-N numbers are determined. Then you can look backwards to see what the gain margin is for consumer audio toys.
Yah, sorry -- it was pretty late here by the time I wrote that. The end of the sentence:
Of course multiple op-amp sections in series are necessary to provide the 10B's functionality. But each one only contributes whatever *excess noise* it contributes right where it is in the circuit. There is no additive, or cumulative, or summing effect such that multiple outputs suffer the increase.
I'd like to look at the schematic! I did vacuum down the PDF 'Owners Manual' offered earlier, but didn't see the schematic. I'll go back and check for others I might have missed.
Pretty sure you won't go wrong following johnmath's advice.
Cheers
should read "that a particular section might contribute more noise to multiple / more than one output."
Of course multiple op-amp sections in series are necessary to provide the 10B's functionality. But each one only contributes whatever *excess noise* it contributes right where it is in the circuit. There is no additive, or cumulative, or summing effect such that multiple outputs suffer the increase.
cheater said:
I'd like to look at the schematic! I did vacuum down the PDF 'Owners Manual' offered earlier, but didn't see the schematic. I'll go back and check for others I might have missed.
Pretty sure you won't go wrong following johnmath's advice.
Cheers