Spectrogram analysis

[Lttlwing16]

Hey folks, I'm putting together a side by side comparison of a couple phono preamps and wanted some input on how to properly interpret the spectral data from audacity.

In particular I'll be looking at System noise, Gain, Record noise and pops.

Here are a couple of spectral samples I've made in audacity. In particular I'm looking at record noise in this set. I've set the Spectrogram settings the same for both using a spectrogram that extends 0-5k hz and also 0-1k hz. I've split the stereo tracks from both preamps and lined up the both left channels and right channels for comparison. The samples were input gain matched to within 0.3 db upon recording.


First, 0-5k

Second, 0-1k hz

I understand the basics of the spectrogram, frequency is plotted on the y axis, and time along the x axis. I understand background speckles represent background noise.

Is record noise more prevalent in the lower frequency, mid range? Would it appear as a deeper and higher concentration of background noise?

Thanks for any input,

David

 

[dc655321]

Given that the spectra of music is often similar to that of (pinkish) noise, how do you propose to distinguish one from the other with this tool?

Are you looking for particular types of noise - snaps, crackles, pops, etc?

 

[Lttlwing16]

Sorry dc, I was inquiring what *could* be gleaned from a spectrogram and how to properly interpret what is shown. For this particular analysis I was exploring: Does one preamp handle record noise better? What frequency range does inherent record noise live at? We know from the review of the PREII by @amirm it had low headroom (22mv input) before clipping. I was exploring if that could be demonstrated in spectral data as well.

I know from comparing the two spectrogram for each channel for each preamp, there is, what appears to me, noticeably more background noise on the PREII and at a higher db vs the MM-6. Could this be accounted for in the .30db difference when the samples were taken? I'd say some, but I don't know to what extent.

One of my favorite LP's (for which this sample was taken) I discovered had somehow gotten a scratch on it and while bad for listening was good for testing pops. However, at looking the pops on the spectrogram, I needed some input on what, if anything can be obtained from the spectral data. Should I zoom in on it, out? I used the waveform data to measure the gain of each pop for comparison and recorded that, however apart from subjective listening, I didn't know what more objective data, from the tools I have, could be obtained from the handling of pops by each preamp.

 

[dc655321]

My point was that it can very challenging to tease apart noise from signal. Ideally, one could deconvolve the two, but that is practically impossible.

EDIT: forgot to add, you may want to look at @pkane's DeltaWave software as another tool.

Please describe your apparatus and method of measurement.

In general, short duration sounds produce a broad spectrum of frequencies, and louder sounds contain more energy (duh!) so will appear darker in the spectrogram.

Play back the tracks in Audacity and note the coincidence of pops/clicks with the cursor's position on the spectrogram. If you note something of interest, you can always zoom in on that region to look for details.

 

[Lttlwing16]

--Oh sure I understand, much of which, I failed to mention above, this sample has next to no music, in fact it may have two single notes played on a piano. And they are all on the left channel. One at around 750hz and 395hz. The rest is record noise. Zoomed in on the music:

This is at the end of a track and the pops can be seen and heard in the dead space between songs:

Apparatus: AT 95ML cartridge>Technics SL-23 turntable> PREAMP (designated in tests)> THX 789 SL INPUT> THX 789 SL OUT> PC soundcard>Audacity 196khz 32bit Float

These samples were taken by alllowing both preamps to remain on for several weeks. If power is cut to change cables, the preamps are allowed ample time to renenergize before measurements are made. Samples were all measured at the same recording level and gain was matched manually by adjusting the PREII gain control until levels being compared were under 0.1 db RMS by measuring within Audacity. Only moving variable is the preamp.

Zooming in on these is definitely has given a closer look at some cool stuff. I'll annotate what I believe I'm seeing in put it in the comparison when I write it up.

 

[Lttlwing16]

In the below is a comparison of the aligned left channel output of two recordings of the same record by the two different preamps.

First, I've outlined what I believe to be areas of groove noise. As one can see, some of these are the same, but some are in different locations, amplitude, and range. This most likely can be explained by the location of artifact moving between the two recordings. Imagine a small piece of dust or dirt being moved by the stylus as it moves along the groove, or a small piece of dust landing on the record which wasn't there in the first recording. As such, it would be hard to come to a conclusion about how each preamp handles these particular noises, since the noise itself is variable. One must also use caution in evaluating other parts of the graph due to these areas affecting the entire frequency range and amplitude at their respective locations.

Now, in the two graphs below we see two notes that appear in the song, and these are the same for both recordings. One is at 750hz and another around 390hz. Here I believe we can see that the left channel on the bottom sample has better resolution. This can be seen by the tighter line at the frequency, with less blooming, and more definition within the pink portions of the graph. The MM-6 has less secondary resonance around the target frequency. Is this audibly significant? I think so. In A/B testing the thing I keep picking up on is resolution.

 

[dc655321]

Here I believe we can see that the left channel on the bottom sample has better resolution

This could be fully characterized by performing a frequency sweep of both devices.

 

[Lttlwing16]

This could be fully characterized by performing a frequency sweep of both devices.

@krichard2496 recommended something similar at 1khz tone, but I unfortunately lack a test record.

 

[AnalogSteph]

Consider buying or making an inverse RIAA circuit, which permits connecting an ordinary high-level output. While you would be bypassing the area of cartridge and cable / input impedance interaction in the high frequencies (hard to characterize without a test record), you would still be able to measure the frequency response of the amplifying circuitry itself with decent accuracy, depending on how accurate the IRIAA is of course.

 

[Blumlein 88]

@krichard2496 recommended something similar at 1khz tone, but I unfortunately lack a test record.

You can do a comparative FR using the two recordings you have. In Audacity do the FFT plot spectrum for the whole track. I wouldn't use one larger than 4096. One option is to export the result. Then do the same for the second recording. You can open the two exported files in a spreadsheet. It will list the value for each bin of the FFT. Just subtract the db value of the 2nd recording from the first for each individual FFT bin. The result will be a differential frequency response.

Or send me the exported FFT files and I'll do it for you.

 

[Lttlwing16]

You can do a comparative FR using the two recordings you have. In Audacity do the FFT plot spectrum for the whole track. I wouldn't use one larger than 4096. One option is to export the result. Then do the same for the second recording. You can open the two exported files in a spreadsheet. It will list the value for each bin of the FFT. Just subtract the db value of the 2nd recording from the first for each individual FFT bin. The result will be a differential frequency response.

Or send me the exported FFT files and I'll do it for you.

Thanks I'll give it a shot this evening and send the FFT files through.

This differential FR will then demonstrate how resolving each preamp is vs one another?

Is the spectral data useful in this regard?

 

[Blumlein 88]

Thanks I'll give it a shot this evening and send the FFT files through.

This differential FR will then demonstrate how resolving each preamp is vs one another?

Is the spectral data useful in this regard?

The differential FR will tell you the Frequency response of one file vs the other. Just looking at the spectral data the second looks to be a little lower in level in the lower frequencies, but that is hard to read of the spectrogram with any precision.

 

[Lttlwing16]

These are the recombined stereo samples of the above spectrograms. As measured, less than .1 dB difference in stereo. I've attached the FFT export.

IF these are two small a sample size, I've attached two other FFT files from a Left Channel analysis of Pearl Jam. (See LT CHAN files below). These measured within .11 dB of each other.

 

[Blumlein 88]

Okay, forgive the goofy formatting, but here are the graphs of the left and then right channel (assuming LTCHAN means left channel).
The left channel looks pretty good, but above 20 khz something is going on in the right channel. Mainly the level was way down in the right channel and gets so low at higher frequencies you are just graphing noise against noise. Y axis is in db and X axis is in hertz. In general in both devices the noise level is 30-40 db lower in the right channel.

 

[Lttlwing16]

Okay, forgive the goofy formatting, but here are the graphs of the left and then right channel (assuming LTCHAN means left channel).
The left channel looks pretty good, but above 20 khz something is going on in the right channel. Mainly the level was way down in the right channel and gets so low at higher frequencies you are just graphing noise against noise. Y axis is in db and X axis is in hertz.

View attachment 148000

Sorry, my fault. I should've labeled them better. The files labeled LTCHAN are from a completely different audio sample than the FFT files.

LTCHAN files: Pearl Jam PreII Lt channel vs Pearl Jam MM6 Lt channel).

The files labeled FFT are from a sample of Local Natives, but are stereo samples for both the PreII and MM6.

 

[Blumlein 88]

I realized I probably didn't understand your original spectrograms. My method of extracting differential FR needs some signal to work with. Playing music for at least 10 seconds or so. You've mostly got a groove noise recording I suppose. And levels are much lower in the right channel vs the left. If you would get the FFT data for 10 ro 30 seconds of music then we'd have something more useful for relative frequency response.

 

[Blumlein 88]

Okay so what I have called the left channel is the LTchan files compared. And the right channel is the other one. Which is probably off because it is stereo.

 

[Lttlwing16]

I realized I probably didn't understand your original spectrograms. My method of extracting differential FR needs some signal to work with. Playing music for at least 10 seconds or so. You've mostly got a groove noise recording I suppose. And levels are much lower in the right channel vs the left. If you would get the FFT data for 10 ro 30 seconds of music then we'd have something more useful for relative frequency response.

Thanks for your help with this.

The files labeled LTCHAN are 36 second samples of full bandwidth music, just from the left channel of each respective preamp.

Did you need the *stereo* file for these 36 second samples?

 

[Lttlwing16]

Okay so what I have called the left channel is the LTchan files compared. And the right channel is the other one. Which is probably off because it is stereo.

Exactly. Sorry about the confusion.

My goals were to evaluate three areas: System Noise, Gain, and Record Noise (groove distortion, clicks, pops) using different samples of music for the Gain evaluation (Pearl Jam) and Record noise (Local Natives). System noise was evaluated by simply recording each system into audacity with nothing playing but energized.

The Local Natives samples (FFT.txt labeled files) were pulled apart and used to evaluate record noise because the left channel had only two single notes across the sample (spectro from post #6) and I figured I could have a good look at groove noise and resolution of those notes. The right channel of the same sample had 4 distinct pops from a scratch on the record, and so it was used to evaluate how each preamp handeled those pops.

From the spectrogram data in post #6 I showed that the groove noise is variable due to the nature of record playback. When the stylus runs across the groove it creates these straight clicks across the entire frequency range, usually at higher amplitude at the lower frequencies, but not always.

If you're not bored to tears -- I'd be happy to grab whatever audio you need when I get home from work this evening to help do a proper differential FR.

 

[Blumlein 88]

Exactly. Sorry about the confusion.

My goals were to evaluate three areas: System Noise, Gain, and Record Noise (groove distortion, clicks, pops) using different samples of music for the Gain evaluation (Pearl Jam) and Record noise (Local Natives). System noise was evaluated by simply recording each system into audacity with nothing playing but energized.

The Local Natives samples (FFT.txt labeled files) were pulled apart and used to evaluate record noise because the left channel had only two single notes across the sample (spectro from post #6) and I figured I could have a good look at groove noise and resolution of those notes. The right channel of the same sample had 4 distinct pops from a scratch on the record, and so it was used to evaluate how each preamp handeled those pops.

From the spectrogram data in post #6 I showed that the groove noise is variable due to the nature of record playback. When the stylus runs across the groove it creates these straight clicks across the entire frequency range, usually at higher amplitude at the lower frequencies, but not always.

If you're not bored to tears -- I'd be happy to grab whatever audio you need when I get home from work this evening to help do a proper differential FR.

Get me 30 or more seconds of full music. If you already have that in the LT Chan files then no need to redo it. While there is some noise the basic FR in the LT Chan files is basically flat and even.