Also could be a USB problem, see https://www.diyaudio.com/community/threads/audio-and-chassis-grounding.427613/page-2#post-8012211 .
THANKS
That is a true "Inside The Box" thinking while many of us were trying to solve the issue "Outside The Box".. I just took a look at my Zenductors... about five feet from my left hand, with their really HOT stacked heatsinks.
Hmm... another project.
Ideas...
For bias.. how about drilling a couple of 1 inch holes on the top lid above the pots? That way you could adjust the bias with the lid off? You could also drive the bias test points to the test points to the outside and use a DMM... this is how my B1K is built. I can adjust and measure the bias from the outside.
Power supplies.... you went to a bigger power supply?
Oh, the back panel is confusing. I'm trying to make sense out of all the connectors. I figured it would have two low level inputs (2 jack), four posts for the speakers and one for the power supply.... but I see more..... I guess I don't understand that... is that to provide power ON/OFF from the front panel?
That is a true "Inside The Box" thinking while many of us were trying to solve the issue "Outside The Box".. I just took a look at my Zenductors... about five feet from my left hand, with their really HOT stacked heatsinks.
Hmm... another project.
Ideas...
For bias.. how about drilling a couple of 1 inch holes on the top lid above the pots? That way you could adjust the bias with the lid off? You could also drive the bias test points to the test points to the outside and use a DMM... this is how my B1K is built. I can adjust and measure the bias from the outside.
Power supplies.... you went to a bigger power supply?
Oh, the back panel is confusing. I'm trying to make sense out of all the connectors. I figured it would have two low level inputs (2 jack), four posts for the speakers and one for the power supply.... but I see more..... I guess I don't understand that... is that to provide power ON/OFF from the front panel?
Last edited:
Thanks for the suggestions, Tony. I could move the power supplies inside the chassis, but then I'd want a proper power inlet with fuse and filter and switch; I like the small DC jacks, and the external supplies are not too much of a hassle.
Regarding bias: I think I'm pretty close to picking a 'final' setting, and will use the built-in DVMs just for monitoring warmup and stability.
I am though going to replace the original 7/8" 6-32 metal standoffs with 20mm M3 nylon ones, to gain a tiny bit (2mm) of clearance between the top lid and the inductors (currently touching, with a kapton strip in between), and I might redo some of the internal DC wiring with RG316 cable. But before I do any rewiring I am going to try the effects of a USB isolator and of a galvanic separation transformer (both of which I already have, but forgot about....) on the noise and distortion measurements. Those bumps and spikes just look very suspicious.
Regarding bias: I think I'm pretty close to picking a 'final' setting, and will use the built-in DVMs just for monitoring warmup and stability.
I am though going to replace the original 7/8" 6-32 metal standoffs with 20mm M3 nylon ones, to gain a tiny bit (2mm) of clearance between the top lid and the inductors (currently touching, with a kapton strip in between), and I might redo some of the internal DC wiring with RG316 cable. But before I do any rewiring I am going to try the effects of a USB isolator and of a galvanic separation transformer (both of which I already have, but forgot about....) on the noise and distortion measurements. Those bumps and spikes just look very suspicious.
Also, regarding 'outside the box': I did and do like the look of the naked amplifiers. Somebody should try a conformal coating and test how that holds up to everyday use, dusting and handling. There are no high voltages here. With a standard coating (25 micron or higher thickness) one probably wants to mask the heatsinks. At work at some point we were thinking about parylene coating of whole modules, but then gave up on it (too hard to get a good estimate on long-term effects and reliability for a large number of devices); that can be done in the sub-micron to tens of micron range, and there are no solvents involved. But most people or even labs don't have a parylene coating reactor/chamber. There are companies that offer it as a service.
I've been moving in circles regarding the noise. I believe what it boils down to is that there are noise spikes at discrete frequencies below -90 dBu, possibly harmonics of 167 Hz (?), and a noise floor that follows a power law (linear in log-log plot) from about -115 dBu around 400 Hz to -100 dB at 20 kHz. Depending on the exact situation (second channel powered up, with or without input signal, input shorted or not, preamp in the loop or not, etc..) the noise floor on the observed ZD2 channel rises 12 to 15 dB and covers up the spikes. There is also some odd behavior of the third harmonic as plotted by REW, with a >10 dB dip around 2 kHz appearing and disappearing, that may or may not be an artefact.
My preliminary conclusion, after close to 100 measurements while trying to systematically vary conditions, is that my setup does not seem conducive to reliable measurements below -90 dBFS.
My preliminary conclusion, after close to 100 measurements while trying to systematically vary conditions, is that my setup does not seem conducive to reliable measurements below -90 dBFS.
Switching to stepped sines in REW seems to get rid of my noise problems (even though I still do not understand the noise floor display normalization), so here are some low bias (1 V) high distortion (> 1%) graphs, made using REW with the Fcousrite Scarlett Solo 3rd gen, the ZD 2 amp pair, and two 8 Ohm 100W resistors in parallel as a 4 Ohm load. (The noise floor rising with frequency in my earlier measurement attempts is apparently a product of the logarithmic frequency sweep in the 'standard' measurement setup.)
Left channel:
The fundamental is off-scale, at 1.2 V nominal input to the ZD2 (1.203 V rms measured for the left channel, 1.194 V for the right channel, at 1 kHz), with about 2.8 V (2.799 V left/ 2.832 V right, again at 1 kHz) appearing at the speaker output, which is about 2 W power into 4 Ohm. The bottom curve is the integrated noise, just below the 7th harmonic. (Noise floor is off-scale below.)
Right channel:
Left channel:
The fundamental is off-scale, at 1.2 V nominal input to the ZD2 (1.203 V rms measured for the left channel, 1.194 V for the right channel, at 1 kHz), with about 2.8 V (2.799 V left/ 2.832 V right, again at 1 kHz) appearing at the speaker output, which is about 2 W power into 4 Ohm. The bottom curve is the integrated noise, just below the 7th harmonic. (Noise floor is off-scale below.)
Right channel:
Last edited:
I did clean up the DC wiring a bit:
After exchanging the metal standoffs for slightly shorter Nylon parts, to gain clearance between the inductors and the top lid, I had to redo a few of the Mosfet wires that had broken off. The RG 316 (inner and outer conductor shorted and used as a single wire) is quite a bit more flexible than the red stranded 12 gauge wire I had been using.
After exchanging the metal standoffs for slightly shorter Nylon parts, to gain clearance between the inductors and the top lid, I had to redo a few of the Mosfet wires that had broken off. The RG 316 (inner and outer conductor shorted and used as a single wire) is quite a bit more flexible than the red stranded 12 gauge wire I had been using.
Last edited:
Your "measured" noise may be a measurement issue or it may be noise from your amplifier.
What is the sensitivity of your speakers? Do you hear noise when your speakers are connected to your amplifier.
When I measure the noise in my amplifiers I input a 1kHz signal to output 2.83Vrms into 8R resistor load (1Watt) and use REW in RTA mode to output a FFT plot. The noise level is referenced to the 2.83Vrms output signal so I can compare noise between amplifiers. I suggest you give that a try.
Here is an example of a FFT plot of one of my amplifiers showing the noise floor referenced to an 1W at 8R output signal:
What is the sensitivity of your speakers? Do you hear noise when your speakers are connected to your amplifier.
When I measure the noise in my amplifiers I input a 1kHz signal to output 2.83Vrms into 8R resistor load (1Watt) and use REW in RTA mode to output a FFT plot. The noise level is referenced to the 2.83Vrms output signal so I can compare noise between amplifiers. I suggest you give that a try.
Here is an example of a FFT plot of one of my amplifiers showing the noise floor referenced to an 1W at 8R output signal:
Hi Ben, thanks for the input! No speakers in the loop for the measurement (but you probably didn't mean that). But when I do connect to speakers (86 dB SPL per W at 1m), I don't think I hear noise, not even with the preamp in the loop and all the way turned up. These days my ears are good to a bit over 12 kHz, I think. The noise floor in the house is about 30 to 35 dB SPL, so I wouldn't really expect to be able to hear noise 80 to 85 dB down from a 'reasonable signal'. The amp would have to be turned up a lot higher.
These (mine and yours) are also very different plots. If you take the signal, harmonics and noise numbers from your distortion box (top left in your graph), and plot them for 3 points per octave from 20 Hz to 20 kHz, you would get the equivalent of my plot, I believe. The signal to noise ratio is about 80 dB in both cases though (my signal is at about 0 dB, noise -80 to -85 dB, your signal is -17 dB and noise -96.4 dB).
So for the above plot you are using the 'RTA' panel with the generator set and turned on from the generator panel, if I understand correctly, and the generator output adjusted so that you measure sqrt(8) Volt at the speaker output. What other settings are you using? If I understand the REW manual correctly, averaging with a shorter FFT width is better (faster for the same noise reduction) than longer FFT without averaging.
Is the noise number in your graph (the -96.4 dBFS) integrated to 20kHz, or to 45 kHz? Either way, the difference is only 3.5 dB, so your noise floor of about -140 dBFS/sqrt(Hz) times sqrt(bandwidth) and your integrated noise as given in the distortion box are close to agreeing with each other (-140 + 43.5 = -96.5, assuming 45 kHz). I should check whether I can now get to that same point. I got really side tracked by trying to understand the noise normalization in the sweep plots, and then by the V/sqrt(Hz) scale in the RTA plot. I'm slowly coming to the conclusion that that is just a waste of time.
These (mine and yours) are also very different plots. If you take the signal, harmonics and noise numbers from your distortion box (top left in your graph), and plot them for 3 points per octave from 20 Hz to 20 kHz, you would get the equivalent of my plot, I believe. The signal to noise ratio is about 80 dB in both cases though (my signal is at about 0 dB, noise -80 to -85 dB, your signal is -17 dB and noise -96.4 dB).
So for the above plot you are using the 'RTA' panel with the generator set and turned on from the generator panel, if I understand correctly, and the generator output adjusted so that you measure sqrt(8) Volt at the speaker output. What other settings are you using? If I understand the REW manual correctly, averaging with a shorter FFT width is better (faster for the same noise reduction) than longer FFT without averaging.
Is the noise number in your graph (the -96.4 dBFS) integrated to 20kHz, or to 45 kHz? Either way, the difference is only 3.5 dB, so your noise floor of about -140 dBFS/sqrt(Hz) times sqrt(bandwidth) and your integrated noise as given in the distortion box are close to agreeing with each other (-140 + 43.5 = -96.5, assuming 45 kHz). I should check whether I can now get to that same point. I got really side tracked by trying to understand the noise normalization in the sweep plots, and then by the V/sqrt(Hz) scale in the RTA plot. I'm slowly coming to the conclusion that that is just a waste of time.
86dB speakers can hide a lot of noise. My speakers are 103dB sensitive so amplifier noise is audible. Here is some information on speaker sensitivity and audible noise:
Another way to measure amplifier noise is with an oscilloscope.
A true RMS Volt meter may or may not accurately measure the noise.
For REW and RTA, I use a stand alone oscillator but you can also use the REW generated 1kHz signal.
I always use the same settings for REW RTA FFTs so that I can compare results. I also mainly use the RTA FFTs, for distortion and noise. I do not usually use REW for frequency sweeps.
As mentioned I use a stand alone 1kHz oscillator. I usually use my multimeter which has True RMS to measure the output voltage at the load resistor, and sometimes I use my oscilloscope. I use the RMS output voltage to calculate the power output: P (Watt) = V x V/R
I like the FFT plot for noise as it shows the noise floor and also noise spikes such as 60Hz AC and its harmonics. In noisy amplifiers the plot may show a high noise floor and also many noise spikes at many frequencies.
Here are my REW distortion settings. The averaging that I use results in a cleaner looking noise floor than if a lower number was used. It allows lower levels of distortion to be visible.
I do not usually pay much attention to the noise numbers in the Distortion box. I just look at the plotted noise relative to the 1kHz 1W at 8 Ohm primary signal level. I like to see the noise floor at least 120dB below the 1W signal (using my REW settings) and also a minimum of noise spikes. I run my computer and oscillator on AC power and that usually adds to the 60Hz and its harmonics so I do not get too excited over that. I do not like to see a noise floor that is not relatively flat and I also do not like to see multiple noise spikes spread throughout the frequency range. If I see that, I go looking for causes.
The amplifier shown in the FFT plot shown in my previous post is quiet when connected to my 103dB speakers - no noise with my ears to the speakers.
Another way to measure amplifier noise is with an oscilloscope.
A true RMS Volt meter may or may not accurately measure the noise.
For REW and RTA, I use a stand alone oscillator but you can also use the REW generated 1kHz signal.
I always use the same settings for REW RTA FFTs so that I can compare results. I also mainly use the RTA FFTs, for distortion and noise. I do not usually use REW for frequency sweeps.
As mentioned I use a stand alone 1kHz oscillator. I usually use my multimeter which has True RMS to measure the output voltage at the load resistor, and sometimes I use my oscilloscope. I use the RMS output voltage to calculate the power output: P (Watt) = V x V/R
I like the FFT plot for noise as it shows the noise floor and also noise spikes such as 60Hz AC and its harmonics. In noisy amplifiers the plot may show a high noise floor and also many noise spikes at many frequencies.
Here are my REW distortion settings. The averaging that I use results in a cleaner looking noise floor than if a lower number was used. It allows lower levels of distortion to be visible.
I do not usually pay much attention to the noise numbers in the Distortion box. I just look at the plotted noise relative to the 1kHz 1W at 8 Ohm primary signal level. I like to see the noise floor at least 120dB below the 1W signal (using my REW settings) and also a minimum of noise spikes. I run my computer and oscillator on AC power and that usually adds to the 60Hz and its harmonics so I do not get too excited over that. I do not like to see a noise floor that is not relatively flat and I also do not like to see multiple noise spikes spread throughout the frequency range. If I see that, I go looking for causes.
The amplifier shown in the FFT plot shown in my previous post is quiet when connected to my 103dB speakers - no noise with my ears to the speakers.
Last edited:
So, in a nutshell you are saying 'forget about the normalization, just figure out where the relative level of acceptable/negligible noise is with a particular FFT & averaging & plotting setting, and then stick to that setting for your own measurements'. I guess in that case I could try to figure out the normalization for that one setting, if I absolutely feel I need to know the exact numbers... ;-)
But that won't help with understanding other people's graphs. I just find it maddening that programs like REW (or Multitone) know the normalization (because how could they compute the integrated noise -in the distortion box- otherwise?), but don't bother to put that information on the graphs by default. This is worse than a measurement without errors (which is not a measurement): it's a measurement without units! Ok, enough ranting. Supposedly the APx500 software does it right, at least if you believe their own note.
But that won't help with understanding other people's graphs. I just find it maddening that programs like REW (or Multitone) know the normalization (because how could they compute the integrated noise -in the distortion box- otherwise?), but don't bother to put that information on the graphs by default. This is worse than a measurement without errors (which is not a measurement): it's a measurement without units! Ok, enough ranting. Supposedly the APx500 software does it right, at least if you believe their own note.