Topic: Measurements Consistency Throughout Different Spectrum Analyzers

Hi,

I am trying to achieve some sort of consistency throughout the software spectrum analyzers that I use in my work, but there seem to be some underlying differences that make them behave differently.

In this example I am using HOFA Analyzer, RME DigiCheck, FabFilter Q3 and Ableton Live’s Analyzer. After some tweaking feeding this setup with a -30dB sine wave results in readings that are consistent throughout the frequency range:

https://i.ibb.co/hKX3BPg/Screenshot-2023-12-15-at-08-36-15.png
100Hz -30dB Sine Wave

*This forum seems to be a bit broken as as per the system message it allows three links per post but it only allows me to post with one link, so please bear with me while I’ll be posting all the supporting evidence.

Re: Measurements Consistency Throughout Different Spectrum Analyzers

https://i.ibb.co/pfbgSZr/Screenshot-2023-12-15-at-08-36-35.png
1000Hz -30dB Sine Wave

https://i.ibb.co/FsBQJQf/Screenshot-2023-12-15-at-08-36-56.png
10000Hz -30dB Sine Wave

Re: Measurements Consistency Throughout Different Spectrum Analyzers

Problems arise with signals that are more complex than a pure sine wave. In the examples below I am feeding it with a -30dB saw wave and you can see that only the three or four first partials are aligned throughout the different analyzers, while the rest of the partials (specifically those of higher frequencies) differ wildly.

https://i.ibb.co/ByBgD88/Screenshot-2023-12-15-at-08-44-46.png
100Hz -30dB Saw Wave

https://i.ibb.co/nBxYNqt/Screenshot-2023-12-15-at-08-45-52.png
1000Hz -30dB Saw Wave

https://i.ibb.co/YhxrnCp/Screenshot-2023-12-15-at-08-46-48.png
10000Hz -30dB Saw Wave

Re: Measurements Consistency Throughout Different Spectrum Analyzers

While I understand that it may be the lack of definition in the higher frequencies where the represented partials must be crumped together tightly, feeding this same setup with a white noise shows how drastically different the same source material may look throughout different analyzers:

https://i.ibb.co/7XZmCrK/Screenshot-2023-12-15-at-08-50-17.png
White noise

Re: Measurements Consistency Throughout Different Spectrum Analyzers

Thanks for reading this far! I am wondering what I am missing here, as I rely on visual information as much as on sonic one and would really like to understand why these behave so drastically different. Would appreciate your help.

Re: Measurements Consistency Throughout Different Spectrum Analyzers

I guess, analysers differ in frequency resolution. Eg. DigiCheck has much fewer bins than others. And they probably differ in time constants, width of data windows, windows overlap function, smoothing results and so....

The more complex signal is, the more those things play the role.... If the source of whitenoise was close to whitenoise, than analysers on the right give much better result, than those on the left...

But it depends what one needs. Analysers on the left may have shorter constants to faster reflect changes in signal.

7

Re: Measurements Consistency Throughout Different Spectrum Analyzers

DigiCheck NG comes with help that explains how its Spectral Analyzer works. The help of the old DIGICheck was even more detailed:

The Spectral Analyser: Theory and Usage
This professional and unique 30 band analyser is based on the Audio Analyser of RME's ADI-96 PRO. After a long time of development we succeeded in transforming its extensive DSP routines into native software code - at a sensational low CPU load and with even highly extended capabilities.

As opposed to nearly all current PC-based solutions no FFT (Fast Fourier Transform) is used. Although this method requires only little calculation (CPU) power, it offers no useable translation of sound into vision. The 'bands' of a FFT have a constant frequency distance, which results in the higher area in numerous bands, in the lower frequencies in very few bands. Some manufacturers try to work around this problem, but even then a sine will show an un-symmetrical display, and an insufficient separation between bands.

RME's Spectral Analyser performes a true bandpass filter calculation, as usual in all professional devices (hardware). The frequency distance between the filters is not linear, but scaled according to human hearing. The highly optimized software allows to run a 30 band analyser with 50 dB range, sharp filters and 100 LEDs per band, without driving the computer to its limits.

Thanks to its complete digital operation the Spectral Analyser offers features previously unavailable from analog analysers. These features include freely adjustable rise and release times, a display configurable in many ways, different filters and special display modes. Tanks to the high calculation accuracy of modern CPUs, the Analyser will show levels precisely down to a jaw dropping -140 dBFS.


Notes on Operation

The most important application using a Spectral Analyser is the visualisation of frequencies and levels found in music or speech. What you see is what your hear! The Analyser shows levels and frequencies even at the edge of the human ear's abilities. The visual display helps to train your ears, and avoids serious mistakes when mixing to the master tape. Usual studio monitors won't let you hear frequencies below 100 Hz. Simply look on the Analyser to see what's going on in the underground!

Reading the display is not easy for novices, because of the huge amount of information that it shows. But after some training you'll agree that this tool is a precious help in every day work.

Analysis of Music

Rise and release time are the most important parameters when trying to achieve a display similar to what you hear. At a very fast rise time setting, the Analyser will show even shortest peaks down to one sample. But these are only audible when present in several bands and lasting for more than a millisecond. That's why a 7 ms rise time results in a much better display behaviour. The release time should be set to three different values. A very fast display (1 s) is recommended for example when trying to see the bass which else hides behind the bass drum. At 2 s the display will be more easier to read and is more friendly to your eyes. Some special applications (like averaging the sound impression or noise measurements) require values around 10 to 20 seconds.

Usually music comes along in stereo. Due to phase cancellations, especially when analyzing signals in SIP mode (Solo in Place), the display does not match the perceived sound. This problem is solved with the mode Max. L/R, which displays the highest level of right and left channel, no matter what their phase relationship is.

Special Display Modes

The Spectral Analyser offers two different envelope modes. Let's look at this example: You want to study the frequency distribution of a bass drum sample.

Method 1: Route the bass drum sample to the analyser and store the display's contents by hitting Freeze in the right moment. Much more convenient is

Method 2: Use Instant Hold to activate the envelope mode. Play the bass drum sample one time. Thanks to the infinite release time the analyser's result won't vanish from the screen.

Method 3 is mainly used with music. At active Instant Hold and display mode Peak Hold the music is shown as usual, but the peak display above the bars won't be reset. When playing back a 3 minute pop song, an envelope of all frequencies and levels which occured in that time is shown.

Sound Measurements

The combination Equalizer/Analyser is used to correct the misalignment of speakers and room. A test signal (pink noise) is send from the speakers into a special measurement microphone placed at the engineer's prefered listening position. Due to the high peaks in noise signals, the Rise time should be set to maximum (4999 ms), Release should be set to 5 s. In Point mode a very precise display of the overall frequency response is achieved. The external Equalizer is now used to correct this curve to a straight horizontal line.

Special Applications

At the highest display range (50 dB) and lowest Top Level setting the Analyser displays all signals down to -140 dBFS. This allows to examine the hum and noise present in the studio's environment. Disturbing frequencies can be classified and removed more exactly.

In Left minus Right mode only the stereo signals within a complete mix are shown. This gives some interesting information on the mixing technique.

Modern (virtual) Music Productions

...often suffer from missing bandwidth limitation and bad programming of PlugIns and (especially) virtual synthesizers. The oversampling mode of e.g. the 4-Bar Level Meter helps to detect hi-frequency components caused by clipping. Of course, the Analyser is a good way to check for overdoses of hi-frequency components as well. Additionally the Biquad filters provide a very useful low-pass filter mode in the lowest band, showing all frequencies down to 0 Hz (DC). Most virtual synthesizers generate inaudible infrawaves, overloading any cutting machine. It's impossible to generate long play records (LPs), as favoured today in dance and electronic music, from such source material. The BiQuad Analyser clearly shows such problematic material. Frequencies below 30 Hz should be removed as good as possible in such a case.

Notes on analysis using noise signals

A FFT shows white noise as horizontal line. This is in most cases also true when examining the noise floor of analog devices, provided those devices do not suffer from hum or other noise signals.

A display based on bandpass filters will show white noise with rising amplitude to higher frequencies, to be more precisely with 3 dB per octave. This has to be taken into account when analysing noise floors of analog devices using the Spectral Analyser.

A linear display requires a noise signal with attenuation of 3 dB/oct to higher frequencies. This signal is called Pink Noise.

But even with such a signal the Rise Time should be set to the highest value, about 5 s. A Release Time set to the same value will cause a display of the average signal, which is well suited for such measurements.


I wrote this text about 20 years ago. FFT based analyzers have much improved since then, but the bandpass still shows music in a better way.

Regards
Matthias Carstens
RME