Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › Too Much Fun with Spectrum Analysis
New Posts  All Forums:Forum Nav:

Too Much Fun with Spectrum Analysis

post #1 of 7
Thread Starter 
Inspiration and Freebies
After reading about the headphone testing at Headroom I wanted to give it a go myself. To keep the upfront costs under control I figured my genuine (biological) "dummy head" would save me the first six grand, and a bit of web surfing might save me the other $15k for the audio analysis gear. After a few false starts I came across a totally stupendous piece of freeware called Spectrogram 5.0 by a guy called Richard Horne. Three thumbs up for this fully functional, intuitive and well-documented program which is only a 244kB download(!)

A laboratory grade noise source would cost another couple of toes so I tuned the FM radio off-station for some white noise to feed the HD580's. Slipping on the 'phones I tucked a $10 stereo tie-clip mic into the left earcup and plugged this via my MD recorder into the soundcard's microphone input. [This soundcard is nowhere near sensitive enough without the extra amplification, and doesn't supply plug-in-power to the mic either].

>>ShShShShShShShShSh<< I'm Testing
Then it was time to let the fun begin before I died of boredom listening to the white noise. I cranked up Spectrogram and selected a high sampling rate, high resolution, scrolling colour spectral display. Wow, so much detail! Almost too much detail, except that the colour codes in the mode I chose were in 5dB steps which was a bit coarse. By gradually reducing the mic gain though, I created a smear graph across time which gave me a highly accurate picture of peaks (horizontal bars of bright colour) and troughs (bars of dark colour). Halting the analysis brings up a cursor which lets you read the frequency and dB response directly off any point in the spectrogram.

No Two are the Same
After noting the main features of the first graph, I switched the mic to the right earcup and produced another smear graph. Needless to say, the results showed similarities but were by no means identical. Had I found a mismatch in my HD580's?? Time to switch back to the left earcup and repeat the analysis. Then I had three different sets of results and a first-hand feeling for the kind of problems involved with this type of testing. It is very difficult to repeat a result if anything has been moved inbetween the tests.

Perhaps the mic shouldn't be in the earcup but rather somewhere inside the ear, which would not be a pleasant experience with this particular microphone. Or should it be at the entrance to the ear without blocking the canal? There must be other factors involved though, since the left and right mic elements are in different locations within the earcup but the spectral agreement between the stereo channels is very good for any single test.

What Does it all Mean?
It's hard to draw conclusions without better repeatability in the tests. One characteristic I noticed on several of the graphs was a dip around 3.7kHz which is where the ear is most sensitive. Presumably this is related to the diffuse field equalisation and may be a key element in a headphone which doesn't cause fatigue. There is another general hole in the response in the 7.5-10.5 kHz region.
Unfortunately though, I have no concrete answer to what I really wanted to know, which is where the airy flutes and violin bowing sounds which characterise the HD580 come from. My best guesses would be the peaks around 4.2kHz, 11-12kHz and 14-15kHz. On the other hand, it could be less to do with dips and peaks and more an issue of consistent phasing(?) I really don't know.

But Wait, That's Not All
There are seemingly endless possibilities with the spectrum analysis software. Tonight I wandered about the place recording all my unfavourite sounds from behind the fridge, TV, computer etc. and plugged them into Spectrogram for a full breakdown of all their annoying component frequencies. Riveting stuff.
post #2 of 7
You are ready to go into business measuring the sounds emitted by rotating machinery and based on the changes predicting failure. It is truly amazing how with a little training you can see a bearing go bad before your eyes.
post #3 of 7
You can use this software to listen ELF/VLF radio frequencies.
for example
82Hz - nuclear submarine communication system....
natural radio etc..
It is really good fun!!!
post #4 of 7
How do you know that the microphone have a linear frequency response?. What you saw could be the microphone.
post #5 of 7
Originally posted by Snusk-Pelle
How do you know that the microphone have a linear frequency response?. What you saw could be the microphone.

That was first thing that come to my mind when I saw the j-curve's post!

To j-curve: be sure to compensate for irregularities in mic response, then the graph will have some value!

post #6 of 7
Thread Starter 
How do you know that the microphone have a linear frequency response?. What you saw could be the microphone.
Too right! But then I get into chicken-and-egg circles... unless I fork out big $$$ for reference test gear. The mic wouldn't explain differences between the left and right earcups or the repeatability problem either. But I agree entirely with your point.
post #7 of 7
Thread Starter 

HD580's signature "gloss" located at 10-10.5 kHz!

I managed to track down where the HD580's airy flute, string and vocal sounds come from. This required a slightly different approach of playing back tracks known to contain lots of HD580 type airy sounds while running the spectrum analyser at the same time (direct from the source). Looking for common frequency content across several glossy notes (of different pitches) narrowed things down to the 10-10.5 kHz region.

An amazing confirmation was made while listening to Mars from Holst's "Planets" suite. (I used the N.Y. Philharmonic, Bernstein, 1973 recording but I think most recordings will bring it out). At 1 minute 11 seconds there is a climax with a screaming piccolo note which really shines through on the HD580. Hopefully the spectrogram is attached below. The vertical axis is frequency (0-20kHz linear scale) and the horizontal is time (about 7 seconds in total). Halfway down the graph there's a thin red line extending to the right. This is the third harmonic of the piccolo note at 10.254 kHz, right in the HD580's sweet spot. You can see the solid red lines of the second harmonic and the fundamental below.

Luckily that kind of piccolo overtone is a pretty rare event at an extraordinarily high frequency, otherwise the HD580 might sound (more) coloured. Most sounds in that frequency range are not single tones but broad spectrum sounds such as cymbals, hissing air and breathing sounds with woodwinds, saxophones and vocals, or the sound of bowed strings etc. I guess it's a colouration no matter how you look at it, just not one which is too offensive on the ear.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Dedicated Source Components
Head-Fi.org › Forums › Equipment Forums › Dedicated Source Components › Too Much Fun with Spectrum Analysis