Does the Real Time Analyzer Block have a Viewing Slope?

ApocalypticKatana

ApocalypticKatana

Power User
Like what you see in Voxengo Span or FabFilter where they have their own spectrum tilt setting…

Does the RTA Block have a tilt applied by default or not? If so what is the db/oct ?
 
It looks to be 3 db per octave when running white or pink noise into it, however when sweeping a sine wave across the full spectrum it looks like it has no slope.

1660362228138.png

1660362550941.png

If the display had a slope, I'd expect those three to be sloped as well. Not sure what's going on there. When I compare in the SPAN plugin with a 3 dB slope the white noise looks like it does above, but the three peaks are sloped as well.

1660363062999.png

1660362999153.png

Here's 500 Hz sine on voice 1, 1kHz sine on voice 2, and white noise on voice 3 compared.
1660363439259.png
1660363465582.png

Span shows slope on both the noise and the sine waves, but the RTA block doesn't. Not sure why.
 
Last edited:
mr_fender said:
It looks to be 3 db per octave when running white or pink noise into it, however when sweeping a sine wave across the full spectrum it looks like it has no slope.

View attachment 106491

View attachment 106493

If the display had a slope, I'd expect those three to be sloped as well. Not sure what's going on there. When I compare in the SPAN plugin with a 3 dB slope the white noise looks like it does above, but the three peaks are sloped as well.

View attachment 106498

View attachment 106497

Here's 500 Hz sine on voice 1, 1kHz sine on voice 2, and white noise on voice 3 compared.
View attachment 106499
View attachment 106500

Span shows slope on both the noise and the sine waves, but the RTA block doesn't. Not sure why.
Click to expand...
I think since the Y db axis on the RTA is separated in 20's visually it's hard to see the 3db/oct

but yeah it does seem to be 3db/oct thanks!
 
mr_fender said:
It looks to be 3 db per octave when running white or pink noise into it, however when sweeping a sine wave across the full spectrum it looks like it has no slope.

View attachment 106491

View attachment 106493

If the display had a slope, I'd expect those three to be sloped as well. Not sure what's going on there. When I compare in the SPAN plugin with a 3 dB slope the white noise looks like it does above, but the three peaks are sloped as well.

View attachment 106498

View attachment 106497

Here's 500 Hz sine on voice 1, 1kHz sine on voice 2, and white noise on voice 3 compared.
View attachment 106499
View attachment 106500

Span shows slope on both the noise and the sine waves, but the RTA block doesn't. Not sure why.
Click to expand...
The RTA block is a "constant Q" analyzer. White noise increases by 3dB/octave. The two tones are equal in amplitude. Therefore you should see a rising slope on the noise and the tones at equal amplitude. I don't know what Span is doing but it doesn't seem right.
 
Ah, I see what Span is doing. They're taking the FFT and then increasing the display by 3dB/oct. This is wrong.

An audio spectrum analyzer is, ideally, a bunch of bandpass filters spaced exponentially in frequency with the Q of each filter constant and proportional to the frequency spacing. A detector is then used at the output of each filter. The reason for the exponential spacing is because that is how humans hear.

No one does it that way though. What you do instead is use an FFT. The problem with the FFT is that it is linear in frequency so your frequency bins are not exponentially spaced. So what you then need to do is combine bins to create the effective bandpass filters. As you go up in frequency you combine more and more bins.

Span is just taking each bin and weighting it by 3dB/octave. You can see it by looking at the noise trace. The perturbations get finer as you go up in frequency.

Voxengo stuff is always weird like this. It's like whoever develops the algorithms sorta knows DSP but doesn't quite understand things fully.

Another way to do it is to use the Chirp-Z transform but this is not as efficient as an FFT and not many people understand how to do that.
 
Related question: what's a good RTA to use with Logic Pro? Not an expert by any means in terms of which ones are precise or not, but I like visually the screenshots I sometimes see posted here with left/right shown overlapped in different colors and differeent frequency bands showing different colors etc.
 
sprint said:
Related question: what's a good RTA to use with Logic Pro? Not an expert by any means in terms of which ones are precise or not, but I like visually the screenshots I sometimes see posted here with left/right shown overlapped in different colors and differeent frequency bands showing different colors etc.
Click to expand...
Waves Paz Analyzer is accurate and should work with Logic Pro.
 
mr_fender said:
Not sure what's going on there.
Click to expand...
I'm not an RTA expert, but in my experience RTA linear displays are raw FFT bin amplitudes. On the other hand, for a bar display (with a constant number of bars per octave) it's necessary to combine bins, and when combining bins you have normalization issues that can give an apparent slope to the display. Linear displays sometimes have a slope parameter to mimic that. In other words, both of your displays are correct, but, by comparing a linear display to a bar display, you're comparing apples to oranges.
 
FractalAudio said:
Ah, I see what Span is doing. They're taking the FFT and then increasing the display by 3dB/oct. This is wrong.

An audio spectrum analyzer is, ideally, a bunch of bandpass filters spaced exponentially in frequency with the Q of each filter constant and proportional to the frequency spacing. A detector is then used at the output of each filter. The reason for the exponential spacing is because that is how humans hear.

No one does it that way though. What you do instead is use an FFT. The problem with the FFT is that it is linear in frequency so your frequency bins are not exponentially spaced. So what you then need to do is combine bins to create the effective bandpass filters. As you go up in frequency you combine more and more bins.

Span is just taking each bin and weighting it by 3dB/octave. You can see it by looking at the noise trace. The perturbations get finer as you go up in frequency.

Voxengo stuff is always weird like this. It's like whoever develops the algorithms sorta knows DSP but doesn't quite understand things fully.

Another way to do it is to use the Chirp-Z transform but this is not as efficient as an FFT and not many people understand how to do that.
Click to expand...

That makes sense. The discrepancy between them was a bit puzzling and I had no idea which was correct. Great info.

I've never understood the point of the variable slope in SPAN. Seems seems like you would always want a "true" display of the frequency content with nothing skewed in any way. It defaults to a fairly high slope as well which confused the hell out of me in the past.

Quick question, why does white noise increase by 3dB? I though it contains frequencies of all equal amplitude. Someday I hope to understand all of this stuff better.
 
mr_fender said:
That makes sense. The discrepancy between them was a bit puzzling and I had no idea which was correct. Great info.

I've never understood the point of the variable slope in SPAN. Seems seems like you would always want a "true" display of the frequency content with nothing skewed in any way. It defaults to a fairly high slope as well which confused the hell out of me in the past.

Quick question, why does white noise increase by 3dB? I though it contains frequencies of all equal amplitude. Someday I hope to understand all of this stuff better.
Click to expand...
White noise spectral density is constant per frequency. However an AUDIO spectrum analyzer is logarithmic. The bandwidth increases with frequency so the power per band in white noise would also increase.

Pink noise spectral density decreases proportionally w/ frequency so the power per band would be constant.

It's not that white noise isn't "flat", it's how we are measuring things.

An analogy would be a sieve. If all the holes are the same size and you pour sand in it the pile below each hole would be the same size. For an audio spectrum analyzer we use a sieve where the holes get bigger and bigger. If you poured sand in that sieve the piles would be proportional to the hose size. If you poured "pink" sand in it the piles would be the same size because the pink sand would have more big grains than small ones.
 
mr_fender said:
I've never understood the point of the variable slope in SPAN. Seems seems like you would always want a "true" display of the frequency content with nothing skewed in any way. It defaults to a fairly high slope as well which confused the hell out of me in the past.
Click to expand...
I think the thing you're missing is the distinction between a linear and a bar display. The point of the slope parameter is to give the linear display in SPAN the slope you're used to seeing in a bar display like the one in the Axe-FX. However, both are correct. As Cliff says, the slope is just an artifact of how the bin data is displayed.
 
GlennO said:
I think the thing you're missing is the distinction between a linear and a bar display. The point of the slope parameter is to give the linear display in SPAN the slope you're used to seeing in a bar display like the one in the Axe-FX. However, both are correct. As Cliff says, the slope is just an artifact of how the bin data is displayed.
Click to expand...
I think the method Span is using is simply wrong. An audio spectrum analyzer should show the power per band with each band being a constant Q (i.e. bandwidth proportional to frequency). The reason for this is because that's how human hearing works. Human auditory perception is akin to a wavelet transform with log frequency spacing. It's NOT like an FFT. We perceive frequency logarithmically. To recreate that visually you want constant Q bands, not linear bins like an FFT produces.

It's not a "linear display". Whether you use the straight FFT data or combining bins to get constant Q it's still a discrete number of points. You can display these points as either a bunch of bars or as a line graph. If using a line graph you can either draw a straight line between points or use some sort of interpolation to smooth the curve. The RTA in the Axe-Fx could also plot a line instead of bars. It's just a different way of displaying the data.

What Span is doing is taking the FFT bins and plotting them as a line graph. To "compensate" for human hearing it is weighting each point proportional to frequency. It's just not the correct approach and results in problems as demonstrated above. The right way, IMO, is to combine bins to create constant-Q bins and plot those bins.
 
FractalAudio said:
I think the method Span is using is simply wrong.
Click to expand...

I agree that a bar display that combines bins into a fixed number of bars per octave serves a purpose. That's why they're popular. However, the slope artifact only occurs when combining bins and a linear display does not combine bins. That's why they give a flat response. It's a discrete number of points, but that number of points is simply the number of bins. Please correct me if I'm wrong, but that's how all linear RTA plugins work, not just SPAN.

The topic of this thread is the relatively unusual feature of SPAN where it has as optional display slope which gives it an weighted appearance that mimics the slope of a bar display. As you say, it's not a true representation of what you'd get if you combined bins. It's just intended to be a rough way to give the display the slope that you're used to seeing.

But my point in responding to Mr. Fender is to explain the discrepancies he asked about, not to take sides on the merits of different display strategies.
 
FractalAudio said:
White noise spectral density is constant per frequency. However an AUDIO spectrum analyzer is logarithmic. The bandwidth increases with frequency so the power per band in white noise would also increase.
Click to expand...
Ok I think I get it. Because the bandwidth increases, there's sort of more overlap in each band so the net power is higher as the bands go up?
 
mr_fender said:
Ok I think I get it. Because the bandwidth increases, there's sort of more overlap in each band so the net power is higher as the bands go up?
Click to expand...
Yes, to keep the number of bars per octave constant, the bandwidth increases and you must combine more and more bins as the frequency increases. But a linear display, like in SPAN, doesn't do that, which is why you saw a difference in the slope behavior.
 
OK that makes sense now. Since the Sine waves are only a single frequency, that overlap is not there so they display flat. Cool.
 
You must log in or register to reply here.
Back
Top Bottom