A good multitone test will stress an audio device in more meaningful ways. Two issues however- first is getting tone sets that don't have harmonics and IM produts on top of other tones. Second is the diagnostic component- figuring out where the circuit needs attention from the frequency domain is difficult at best. The residual from a notch can reveal a lot about where in the cycle and where in the circuit the non-linearity lies. I'm attaching some info on "Spectral Contamination" that may be helpful.
Demian,
Thanks, from all of us!
I've been using the "NID" multitone test pattern as generated in REW for some simulations in LTspice. It's really interesting how some preamp and amp designs look pretty similar when simulated for harmonic distortion yet can be 10 dB or even 20 dB apart in the multitone performance.
But, I'll try these.
Thanks, from all of us!
I've been using the "NID" multitone test pattern as generated in REW for some simulations in LTspice. It's really interesting how some preamp and amp designs look pretty similar when simulated for harmonic distortion yet can be 10 dB or even 20 dB apart in the multitone performance.
But, I'll try these.
OK, I think I got it right.
This is a simulation of the simplest circuit I can think of. Two 2SK170BLs run at Idss, with one as the current source and one as the buffer. 15 VDC rails. 1 Vrms total input voltage to 10K load.
I won't belabor this, because it's the wrong forum thread, but wanted to show an example of the Jon Risch Phi 12 multitone test. The floor doesn't include any circuit noise - only calculation "noise".
This is a simulation of the simplest circuit I can think of. Two 2SK170BLs run at Idss, with one as the current source and one as the buffer. 15 VDC rails. 1 Vrms total input voltage to 10K load.
I won't belabor this, because it's the wrong forum thread, but wanted to show an example of the Jon Risch Phi 12 multitone test. The floor doesn't include any circuit noise - only calculation "noise".
Hello,
Here is a Cosmos 9039S -> Cosmos ADC measurement. I hope this is what you wanted. 9039S is at 0dBFS (3.5Vrms)
Last edited:
I'm still at that stage where the more I use REW, the less I find I understand what it does, in particular with regard to graphing noise in a spectrum type plot (sqrt(power)-quantity vs frequency) from an FFT. I tried to describe my question in another thread, but it would also fit here, I believe: https://www.diyaudio.com/community/threads/noise-and-fft-condundrum.346959/#post-8016171 . I'd appreciate any hints or clarifications. I have consulted the REW manual, and also https://www.ap.com/news/fft-scaling-for-noise .
Yes, most likely.
Unless there is some distortion cancellation here. If the DAC and ADC have distortion of the same magnitude, distortion cancellation can happen if the phase of the two are different.
The multitone distortion pattern that Demian suggested is very easy to implement in Multitone. (Oddly enough)
The above is a Topping E50 DAC (balanced output) > E1DA Cosmos Scaler > E1DA Cosmos ADCiso, with the ADCiso's attenuator set to an input level of 4.5 Vrms. As Jon Risch suggested in his paper, I ran the DAC at -3 dBFS.
This is just a representative measurement for the test - not a reference or anything special.
The above is a Topping E50 DAC (balanced output) > E1DA Cosmos Scaler > E1DA Cosmos ADCiso, with the ADCiso's attenuator set to an input level of 4.5 Vrms. As Jon Risch suggested in his paper, I ran the DAC at -3 dBFS.
This is just a representative measurement for the test - not a reference or anything special.
Those are beyond exceptional. Are they analog? I don't see evidence of any power line harmonics even.
A few pointers to ensure the most accurate results. First make sure the source device is isolated from the ADC and does not St at a clock or phase noise/jitter will be cancelled. Second make sure none of the tones are subharmonics of the sample clock. At ensures you use all the bits. . .
A few pointers to ensure the most accurate results. First make sure the source device is isolated from the ADC and does not St at a clock or phase noise/jitter will be cancelled. Second make sure none of the tones are subharmonics of the sample clock. At ensures you use all the bits. . .
The test tones were generated by Multitone on Rob Baum's stated frequencies from his spreadsheet for Phi12r - OK, I rounded them to the nearest Hertz. These were played through a Topping E50 DAC, powered by a battery. The balanced output of the DAC was applied to the E1DA Cosmos Scaler, also powered by its own battery. From the Scaler, the signal was then fed into the E1DA Cosmos ADCiso ADC, which was powered by a linear regulated wall wart from Jameco.
The DAC was fed from a USB hub attached to the computer (an Apple Intel iMac running Windows 10 via Base Camp). The output of the ADCiso went through the same hub back to the computer.
The 100 Hz tone Rob Baum specified is a subharmonic of the 192K sampling rate. So, I just tried moving it from 100 Hz down to 97 Hz. All the other conditions are the same. Here are the results. One has a logarithmic display of frequency, the other is linear. Just for demo.
I suspect the slight differences from the previous results may be due to the test set-up not being the same temperature as last time. Couldn't help it.
The DAC was fed from a USB hub attached to the computer (an Apple Intel iMac running Windows 10 via Base Camp). The output of the ADCiso went through the same hub back to the computer.
The 100 Hz tone Rob Baum specified is a subharmonic of the 192K sampling rate. So, I just tried moving it from 100 Hz down to 97 Hz. All the other conditions are the same. Here are the results. One has a logarithmic display of frequency, the other is linear. Just for demo.
I suspect the slight differences from the previous results may be due to the test set-up not being the same temperature as last time. Couldn't help it.
Last edited:
Is there a 32 bit version of the software? For hardware compatibility my primary measurement system runs 32 bit Win 10.
Why does the version of the OS matter? Just curious. It seems you're dealing with data, the OS shouldn't care as long as it can handle with it. Driver issues?
So what is the noise normalization for these spectra? It's not dB/Sqrt(Hz) for one Hz wide bins, because then the level (between the signal peaks) should be at -120 dB (integrated noise) - 43.3 dB (20*log(Sqrt(43,000)) = -163 dB. The displayed level looks more like -150 dB, maybe -148 dB. And that is after the noise reduction due to averaging 32 samples, which reduced the noise already by 15 dB.
Hi Demian,
Okay, makes sense. I agree, if you have to pay for new software it makes less than no sense.
Faster computers off-lease are cheap though (I never buy brand new machines). Stuff them with more memory and SSD drives and you're golden. Typically I install a new video card, or used that I clean all the dust out of. Sometimes a new CPU fan and power supply to support the new hardware. These are the failure points anyway, or bottlenecks. Used machines are typically off-lease business machines with minimal configurations. I get >10 years out of an off-lease machine. You just have to blow them out every year (like a new one). This is an i5 Lenovo, maxed out. It runs my entire bench and everything else I typically do. Dual 24" monitors so I can view a schematic and PCB design at the same time. Monitors are also off-lease, Lenovo. I have an i7 Lenovo box running Fedora. I need to migrate over to that.
Okay, makes sense. I agree, if you have to pay for new software it makes less than no sense.
Faster computers off-lease are cheap though (I never buy brand new machines). Stuff them with more memory and SSD drives and you're golden. Typically I install a new video card, or used that I clean all the dust out of. Sometimes a new CPU fan and power supply to support the new hardware. These are the failure points anyway, or bottlenecks. Used machines are typically off-lease business machines with minimal configurations. I get >10 years out of an off-lease machine. You just have to blow them out every year (like a new one). This is an i5 Lenovo, maxed out. It runs my entire bench and everything else I typically do. Dual 24" monitors so I can view a schematic and PCB design at the same time. Monitors are also off-lease, Lenovo. I have an i7 Lenovo box running Fedora. I need to migrate over to that.
Its drivers for the usb scopes and the AP sys 1. Windows drivers are a real hardware limitation for older stuff. A new AP and new USB scopes because of drivers is an offense to my nature. I have 10 64 bit I7 pcs on hand. Win 11 doesn't have a 32 bit version. But what I have is close enough to SOTA that a 40 dB cost increase to upgrade makes no sense.
I can't agree with you more.
I have some great equipment that ran on the printer port. Enough said. Some had drivers to WinXP, compatibility mode in later Windoze releases do not help. So both those classes of equipment are dead weight, some ran on DOS only. Interface cards that are 8 or 16 bit are likewise dead.
Guess I am dating myself a bit! lol!
I have some great equipment that ran on the printer port. Enough said. Some had drivers to WinXP, compatibility mode in later Windoze releases do not help. So both those classes of equipment are dead weight, some ran on DOS only. Interface cards that are 8 or 16 bit are likewise dead.
Guess I am dating myself a bit! lol!