Understanding a sound spectrum analyzer output

[soupybionics]

Hi,

I am just trying to understand the output of a sound spectrum analyzer. There are three dimensions
1) Time
2) Frequency
3) Volume (db)

I am confused about few things here:
1) How should I perceive the volume dimension here?
2) Why are the db values negative here and w.r.t what reference value are they negative? The standard equation for the volume (sound) is 20(log(value/refvalue) to base 10), right?
3) Also I don't really understand the reason behind multiplying the log output with 20 here. What does it signify?

I have attached the spectrum analyzer output.

Thanks in advance!

 

[davenn]

1) How should I perceive the volume dimension here?

the colour, refer to the colour bar down the right side
most of the largest amplitudes are 4kHz and below

2) Why are the db values negative here and w.r.t what reference value are they negative?

maybe 0 dBuV ?? ... 0 dBV ?? no info is given
if you are using a sound card for this... maybe there are spec's for the card you can find ?
or spec's for the software when related to a particular card ?

log 10 for Watts ( power gain/loss)
Log 20 for Voltage gain/loss

http://www.sengpielaudio.com/calculatorVoltagePower.htmcheers
Dave

 

[soupybionics]

Thanks for the info. It's just a software, there aren't any specs. Could those decibel values be w.r.t the highest amplitude in the song?

 

[jtbell]

The audio processing software that I use (Audacity) defines its 0 db level as the maximum possible amplitude that can be represented, i.e. with all bits in a digital sample set to 1. For 16-bit samples, the floor (all bits zero) is -96 dB. For 24 bits, the floor is -145 dB.

Your spectrogram amplitude scale shows values ranging from -20 dB down to about -116 dB. That's a range of 96 dB, the same as I get with 16-bit samples. So maybe your software simply sets the maximum amplitude arbitrarily to -20 dB.

 

[soupybionics]

Thanks jtbell. When you say "For 16-bit samples", are you referring to values in Time Domain (Voltage) or Frequency Domain?

 

[jtbell]

When you say "For 16-bit samples", are you referring to values in Time Domain (Voltage) or Frequency Domain?

That would be the time domain. The place where I actually see 0 dB is in Audacity's input-level meter. If the input signal is too strong, the waveform clips at that value. Audacity has a spectrogram display similar to yours, with a different color palette. I'd never looked before for the exact correspondence between color and amplitude, but I've done it just now. According to this page:

http://manual.audacityteam.org/o/man/spectrogram_view.html

Audacity applies a gain factor to the signal before translating the amplitudes to colors. The default gain is (surprise!) 20 dB, which effectively sets the "maximum" color to a real signal level of -20 dB, in agreement with your software. I don't know if 20 dB gain is a conventional standard default for this type of display, or if these two programs just happen to use the same value.

 

[soupybionics]

I see. Thanks. So to increase/decrease the volume of a sound , we don't really need the frequency domain data?
Time Domain (Voltage levels for amplitude) data is enough?
Also, I didn't get the deal with applying the gain factor . How does that translate to a maximum of negative 20db?

 

[sophiecentaur]

So to increase/decrease the volume of a sound , we don't really need the frequency domain data?

It's the same signal, however you choose to describe it. You can either scale all the time samples by a 'gain' factor or do the same with all the frequency domain values. Both methods give you the same resulting signal.

 

[nasu]

I see. Thanks. So to increase/decrease the volume of a sound , we don't really need the frequency domain data?
Time Domain (Voltage levels for amplitude) data is enough?

You don't really need any data to physically reduce the volume of the sound.
You just put some sound absorbing material in the path of the sound wave or just turn down the power of the source, if possible.
But maybe you are asking something else?

 

[sophiecentaur]

The dB scale is a great one for causing confusion and disagreement between people who are unfamiliar with it. It doesn't help when people try to help you by telling it wrong, too. (I am not pointing any finger but you all will know what I mean.)

How does that translate to a maximum of negative 20db?

This is another issue, to do with the use of the dB scale. the deciBel is a ratio of powers and it can either be used as just that - e.g. 'increasing the output power of a 10W signal by 20dB will give you 1000W. When measuring quantities in dB, you have to say what power you are comparing it with - e.g. "the power was -6dbW, meaning the power was 1/4W. The dB is a very versatile measure because it covers a vast range of values with only a few digits - +60dB is the same as 1000000 times the power.
In measuring instruments, it is common to use the mW as the reference value in audio because audio distribution systems (and telephones) used to use 600Ω as a standard load value and 1mW is 0.77V (just under a convenient 1v) A dB graph scale can show a huge range of values (to a limited accuracy, of course) much better than a linear scale which can show no better than a range of 1 to 100 units on a scope screen. The top line on the Spectrum Analyser corresponds to -20dB. If you increase the gain of the channel, the top line could be -30dB or 40dB but then some components could be over the top of the screen.

we don't really need the frequency domain data?

People were designing and using electronics systems years before the spectrum analyser existed as a measuring instrument. It is not essential but can be very handy. The frequency domain has been used and understood since Mr Fourier's day, of course, and filter responses have been described in frequency terms since the year dot. But nowadays, we frequently use filters, based on the time domain.

 

[soupybionics]

Thanks for the detailed explanation, sophiecentaur.