I got the idea from reading the specs that the Fluke is flat - which Fluke defines as 2% - to 20 KHz. But, I like your number better. 🙂
So, I'll spend some time tomorrow testing to see just what it really is.
Question: to compare the noise floor in the plots above to other FFT graphs, I need to take into account the FFT bandwidth, the windowing function, and any averaging that might have been done, right?
Longer version of the same question:
According to the REW help file, locking the generator frequencies to the length of the FFT window assures that all the generated signals are periodic in the FFT, so that no signal leakage into neighboring FFT bins occurs (a pure sine will show up at a single frequency in the FFT graph). This also means that you can use a rectangular (= no) window, which is nice for noise calibration, because the 'window factor' is 1.
But I think you might still want to use a non-rectangular FFT window, because distortion products (sums and differences of base frequencies and of harmonics) will not generally be periodic in the FFT length, right?
If you use a (non-uniform) window then you will have to correct the noise floor. I guess you could make two FFT transforms, one with your window function of choice optimized for finding non-frequency-aligned peaks, and the other with no window function (i.e. rectangular window). Or you could just take the noise integral computed by REW, divide it by the square root of the bandwidth, and draw the noise line (assuming it is a flat line, i.e. white noise) by hand.
So, what is the correction factor for the noise floor in the plots above by @CG ?
Longer version of the same question:
According to the REW help file, locking the generator frequencies to the length of the FFT window assures that all the generated signals are periodic in the FFT, so that no signal leakage into neighboring FFT bins occurs (a pure sine will show up at a single frequency in the FFT graph). This also means that you can use a rectangular (= no) window, which is nice for noise calibration, because the 'window factor' is 1.
But I think you might still want to use a non-rectangular FFT window, because distortion products (sums and differences of base frequencies and of harmonics) will not generally be periodic in the FFT length, right?
If you use a (non-uniform) window then you will have to correct the noise floor. I guess you could make two FFT transforms, one with your window function of choice optimized for finding non-frequency-aligned peaks, and the other with no window function (i.e. rectangular window). Or you could just take the noise integral computed by REW, divide it by the square root of the bandwidth, and draw the noise line (assuming it is a flat line, i.e. white noise) by hand.
So, what is the correction factor for the noise floor in the plots above by @CG ?
Should be interesting. I certified many of these meters, and they always performed much better than expected. Mind you, HF performance is very sensitive to the internal shield positions. I took a dead 87 and drilled the case to line up with the adjustment points. The HF readings still changed when I put the original case back on, but they were much, much closer to where I adjusted it to.
With that in mind, your meter may not be close to where it should be, or is capable of. With this type of accuracy, "closed case calibration" was the only real way to maintain the potential of these meters. Sadly these are manual, open case.
I can't tell you how frustrating it was to adjust one of these really close, only to button it up and it might be out of tolerance. That's why I made the drilled case. Not perfect, way better.
With that in mind, your meter may not be close to where it should be, or is capable of. With this type of accuracy, "closed case calibration" was the only real way to maintain the potential of these meters. Sadly these are manual, open case.
I can't tell you how frustrating it was to adjust one of these really close, only to button it up and it might be out of tolerance. That's why I made the drilled case. Not perfect, way better.
Hi Gruesome,
Many meters are far worse. In fact, you would be shocked by how many are not in tolerance brand new out of the box. Then there is the question of time, and holding calibration.
I only buy HP / Agilent / Keysight or Fluke handheld meters. Experiences in the cal lab.
Many meters are far worse. In fact, you would be shocked by how many are not in tolerance brand new out of the box. Then there is the question of time, and holding calibration.
I only buy HP / Agilent / Keysight or Fluke handheld meters. Experiences in the cal lab.
You are completely right.
I set the sample rate in Windows and REW to 384 KHz and plotted the response of the DAC combined with the ADC. The response was down about 2 dB at 170 KHz before the expected crash closer to the filter cutoff at half the sample rate.
Then I used REW and the DAC as a signal generator and methodically collected the Fluke 87 reading with the generator set to 1.00 Vrms for over 100 frequencies. I entered all the data into a spreadsheet and was going make a nice graph of the response. But, I found that the Fluke pretty much tracked the DAC > ADC measurement out to 180 KHz. I mean, within about 0.2 dB. Good enough for audio work.
So, I guess that the earlier results from REW and Multitone were comparable for level.
Hi CG,
Good to know, thanks for measuring this.
It's nice to know an old man's memory isn't faulty either! I just had a good feel from certifying them, we never did bother measuring the response as long as it met spec.
The Fluke 87 is a classic, industry standard meter. Perfect for hobbyists since there were may sold over many, many years. They are well worth repairing and certifying as they do hold their calibration well. The same can be said for the HP 974A, but not many of those were sold. The Fluke 83 and 85 would be very similar, grab them if you see them up for sale and you need something you can trust.
Good to know, thanks for measuring this.
It's nice to know an old man's memory isn't faulty either! I just had a good feel from certifying them, we never did bother measuring the response as long as it met spec.
The Fluke 87 is a classic, industry standard meter. Perfect for hobbyists since there were may sold over many, many years. They are well worth repairing and certifying as they do hold their calibration well. The same can be said for the HP 974A, but not many of those were sold. The Fluke 83 and 85 would be very similar, grab them if you see them up for sale and you need something you can trust.
Regarding noise floor only, switch Y axis scale to V/rt(Hz) - voltage noise density. This setting eliminates all the above mentioned variables and is the only correct one for this purpose. However, can be used only to read noise floor, the noise density. Not the amplitude of harmonics. Remember that total noise would be noise density x root of BW.
I bought this Fluke 87 back in the early 90's. Aside from the switch contact pad update and the better replacement screen from the factory, I've had to do nothing to it aside from replacing the battery. (I now use lithium batteries in all these pieces of electronic gear to avoid the problem of battery leakage.) I can't complain one bit about the service it's provided.
Duracell batteries are the problem children. Everyready batteries typically do not leak.
If I see a Duracell in anything, it gets recycled immediately. I've had Duracells measure like new while leaking. Throw them out !!!!
If I see a Duracell in anything, it gets recycled immediately. I've had Duracells measure like new while leaking. Throw them out !!!!
I just checked a Fluke 8060A, 8050A 87 mkIII and a Soar. The 8060A is .3% low at 100 KHz and 2% at 300 KHz. The 87 mk3 seems to start rolling off around 60 KHz (-1%) and is down about 3% at 100 KHz. The 8050 down 1% at 80KHz and 2.5% at 100KHz. The 8060A is spec'd at 3% to 100 KHz The 87 was last calibrated at Fluke in 2006. And the 3% is more that within spec (max 20 KHz +/- 3%). The 8050 is specified to 50 KHz at +/- 5%. The SOAR (vintage cheapy handheld DVM) just makes to to 10 KHz and drops to almost noting at 20 KHz. I'm using an Optimation AC calibrator that is .01% to almost 1 MHz (per the fluke thermal converter).
The down side of TRMS meters is that they do not show crest factor. For nonsiusoidal waveforms it can be significant, even powerline harmonics can have significant crest factor. With non harmonically related tones they will go through a peak level every so often (related to the least common denominator?) so probably more like random noise. You may need a lot of averages to get all the relationships.
It would be interesting to see if you can duplicate the REW with the IM products. And maybe share your files or settings so we can attempt to duplicate them. There may be intereting things to learn.
The down side of TRMS meters is that they do not show crest factor. For nonsiusoidal waveforms it can be significant, even powerline harmonics can have significant crest factor. With non harmonically related tones they will go through a peak level every so often (related to the least common denominator?) so probably more like random noise. You may need a lot of averages to get all the relationships.
It would be interesting to see if you can duplicate the REW with the IM products. And maybe share your files or settings so we can attempt to duplicate them. There may be intereting things to learn.
For that reason I'll often use a scope to see exactly what I am measuring. Crest factor is a big thing. Also on a DC measurement, just measuring DC volts doesn't mean it doesn't have spikes or dropouts. A true RMS measurement doesn't show you any of that, or the combination displays where some meters will show both DC voltage and the AC component.
Over time you develop that thing where you don't take anything at face value.
Over time you develop that thing where you don't take anything at face value.
I visited a factory in China that looked like a scene out of Dickens. They made 80% of the Alkaline 9V batteries in the world. And had samples with both Everready and Duracell labels. They all leak. I just pitched a bath of new button cells that were leaking. Lithuim usually doesn't leak. Much too reactive to risk leaking potential.
Hi Demian,
I haven't had any AA, AAA or 9V Eveready batteries leak that were not dead, dead, dead. I have Duracell 9V and AA that leaked while the battery indicator on the meter said they were fine (they weren't that old either). One set nearly destroyed an HP 974A. Duracell told me you have to discard their batteries before the date stamped on them. Really?
I'm sorry, but in my day to day experience, and lab experience, Duracell batteries are about the worst you can buy. Even Panasonic batteries are better for leakage.
My advise to anyone who cares ... if you have anything with Duracells in them, discard those batteries right away. Do not buy them either, although they are usually on sale or less expensive. I have got in countless remote controls and other things in service with rotted Duracells in them.
I have seen the odd Eveready with fuzz or leaky a bit. They were dead first. Duracell batteries leak a lot, and they may still have a charge, not full but over a volt. That shouldn't happen unless the cell is very old and dead.
I haven't had any AA, AAA or 9V Eveready batteries leak that were not dead, dead, dead. I have Duracell 9V and AA that leaked while the battery indicator on the meter said they were fine (they weren't that old either). One set nearly destroyed an HP 974A. Duracell told me you have to discard their batteries before the date stamped on them. Really?
I'm sorry, but in my day to day experience, and lab experience, Duracell batteries are about the worst you can buy. Even Panasonic batteries are better for leakage.
My advise to anyone who cares ... if you have anything with Duracells in them, discard those batteries right away. Do not buy them either, although they are usually on sale or less expensive. I have got in countless remote controls and other things in service with rotted Duracells in them.
I have seen the odd Eveready with fuzz or leaky a bit. They were dead first. Duracell batteries leak a lot, and they may still have a charge, not full but over a volt. That shouldn't happen unless the cell is very old and dead.
I think I know what you're asking, and I think I did that back in post 238 Both REW and Multitone used pretty much the same tone frequencies, at least as close as I could make them, and both had an RMS voltage of 0.71 V.
John Mulcahy - author of REW - has said that crest factor optimization algorithms generally have little effect on these kinds of tone sequences. So, I just tried turning crest factor optimization off in both programs and testing again. The RMS output voltages didn't change, and the stated crest factors in each generator didn't change. Results for Total Distortion + Noise stayed the same and were within a dB between the software applications. I get that much variation at this distortion level between tests run 10 minutes apart, no matter the software. Perhaps Audio Precision results are more stable than that, because I see SINAD ratings competitions separated by tenths of a dB on some web sites. 🤷♂️
Oops! Forgot the tone frequencies. I requested REW to generate a sequence beginning at 20 Hz and stopping at 24 KHz. It gave me a starting frequency of of 20.5 Hz. Using that as a base with the sequence John Mulcahy published, the frequencies I also used in Multitone were:
20.5
102.5
266.5
594.5
1004.5
1660.5
2439.5
2890.5
4243.5
5391.5
7359.5
9409.5
11131.5
14944.5
15887.5
18634.5
22488.5
Both applications were set to use 1 Meg FFT lengths, Blackman-Harris 87.5% filters, and 16 averages.
Please correct anything I am doing wrong. Anybody. I'm not proud. No need to be gentle.
Last edited:
Thanks, PMA! So you could pick a non-rectangular window and set up some averaging, and the noise floor in V/sqrt(Hz) would still come out correct? I was under the impression you'd have to also pick the rectangular window and either forego or correct for averaging. I got the correct scaling when I did that. I'll have to check whether window and averaging are automatically corrected next time.
My question though is with regard to measurement plots made by other people, where I can't influence the settings used. For example the plots posted by CG in this thread. Assuming he used the same parameters as in post #255, the FFT window length was 1 million, bandwidth 24 kHz, 3-term Blackman-Harris window (I could not find a reference for an 87.5% Blackman-Harris window, so I'm guessing 3-term), and 16 averages, would I take the noise level indicated in the plot, divide it by the FFT bin width in Sqrt(Hz) (in the example sqrt(24,000/1,000,000), divide it by the square root of the noise power bandwidth (from this AP note, Sqrt(1.73) for Blackman-Harris 3-term), and multiply with the square root of the number of averages (4 in this example)? So in this example it would make for -140 dB (from the most recent plot in this thread) - (-16.2 dB) -2.4 dB + 12 dB = -114.2 dB/Sqrt(Hz)?
But, the Multitone plot indicates an integrated noise value of -120 dB (for the same plotted noise floor level of -140 dB as in the REW graph). Adding 43.8 dB (i.e. multiplying the flat noise floor density with sqrt(24 kHz) bandwidth) to either number for the noise floor, this doesn't seem to make any sense at all, with or without the corrections I just listed.
I would really appreciate if somebody could point out where I'm going wrong; no doubt there is some very basic mistake there.
Last edited:
At this stage I would be looking at EMI/RFI resistance, Settling time, Multitone IM (of course), implementation issues like system grounding on electronics but this all pales compared to speaker issues. Put the time and energy into addressing room and speaker issues. Nothing in my experience has made nearly as much difference as resolving room issues. Those are really difficult- everything from construction to domestic tolerance need to be resolved. And even a perfect speaker won't give the same percieved experience the mixing engineer or mastering engineer had and none of those will be "right".
... And what control room in a studio is acoustically "right"? Answer : Not a one.
I agree Demian, people can't hear stuff below -100dB typically. But the stuff you see down there does in fact point to problems equipment may have in other areas that may be audible under some conditions. Once you get higher currents flowing, layout and grounding become far more critical, and this is where some products misbehave.
I still see a lot of value in paying attention to what goes on in the noise floor. One thing for sure, if problems are seen up to the -80 dB mark, some people can detect them. For sure an amplifier (say) that shows distortion or other noise up to those levels, it simply does not sound as good as an equivalent amplifier that hasn't got those problems.
As I've pointed out before, your listening environment probably hasn't got near the dynamic range of decent equipment. Speaker distortion will (hopefully) dominate, it is different to distortion in your electronics.
I agree Demian, people can't hear stuff below -100dB typically. But the stuff you see down there does in fact point to problems equipment may have in other areas that may be audible under some conditions. Once you get higher currents flowing, layout and grounding become far more critical, and this is where some products misbehave.
I still see a lot of value in paying attention to what goes on in the noise floor. One thing for sure, if problems are seen up to the -80 dB mark, some people can detect them. For sure an amplifier (say) that shows distortion or other noise up to those levels, it simply does not sound as good as an equivalent amplifier that hasn't got those problems.
As I've pointed out before, your listening environment probably hasn't got near the dynamic range of decent equipment. Speaker distortion will (hopefully) dominate, it is different to distortion in your electronics.