Peak Hold in a Display

trogluddite · 24 Oct 2011

Hi, I assume that you want to find the max of a stream signal (changing at sample rate, like audio) - so you need to get the max using stream (blue) parts. The simplest way, is to use an arrangement like this...

When 'On/Off' = TRUE. The 'Max' primitive (top one) here, will compare the present input value with the previous output value and keep the biggest one. This works because when you have feedback from output back to input, the feedback is always the result from the previous sample. Also Boolean TRUE = 1 so the output value is not changed.

When On/Off = FALSE. FALSE = 0, so the max now finds the biggest of the input value and zero --> just the input value. So now the peak is reset.

If you are looking for peaks in an audio signal, remember that sometimes the negative (downwards) part of the audio might be the biggest - so you can add a 'stream abs' to the main input to find only the size of peak, even if it is a negative peak.

Mo · 24 Oct 2011

Nice solution Trog. I think it will be better to make this in green for a display. Also stream abs would use a lot cpu. I tried but it turned red, so found another solution for green:

Here build into the Mono Readout:

trogluddite · 24 Oct 2011

Mo wrote: think it will be better to make this in green for a display

The problem with using the green method, is that the M2F can only trigger at max.100Hz (25Hz for your schematic) - so it is unlikely that the maxima that you find will truly capture the peaks of the audio. To guarantee a true maximum it is necessary to find the peaks in stream form before converting to green.

Mo wrote: Also stream abs would use a lot cpu

True, the stock abs function is not compiled as efficiently as it could be - there is a much quicker way if written directly in assembly, using bitmasking to remove the sign bit...

streamin in; streamout out; int absMask = 2147483647; //largest positive 32bit integer

movaps xmm0,in; andps xmm0,absMask; //sign bit to zero movaps out,xmm0;

Mo · 24 Oct 2011

trogluddite wrote: The problem with using the green method, is that the M2F can only trigger at max.100Hz (25Hz for your schematic) - so it is unlikely that the maxima that you find will truly capture the peaks of the audio. To guarantee a true maximum it is necessary to find the peaks in stream form before converting to green.

Ah yes. Too slow for analyse most audio. trogluddite wrote: there is a much quicker way if written directly in assembly, using bitmasking to remove the sign bit...

Thanks for reminding me, it is a cmpps, a sub and a add less than the new abs code function.

infuzion · 24 Oct 2011

trogluddite wrote: Mo wrote: think it will be better to make this in green for a display

The problem with using the green method, is that the M2F can only trigger at max.100Hz (25Hz for your schematic) - so it is unlikely that the maxima that you find will truly capture the peaks of the audio. To guarantee a true maximum it is necessary to find the peaks in stream form before converting to green.

Mo wrote: Also stream abs would use a lot cpu

True, the stock abs function is not compiled as efficiently as it could be - there is a much quicker way if written directly in assembly, using bitmasking to remove the sign bit...

streamin in; streamout out; int absMask = 2147483647; //largest positive 32bit integer

movaps xmm0,in; andps xmm0,absMask; //sign bit to zero movaps out,xmm0;

I agree, but you have to reset the held max value after it is drawn. This "Peak Hold" is exactly why I made this post: Green 2 Stream - Missed Triggers

tor · 24 Oct 2011

To get it displayed you still need to use the M2F. If there only a few samples exceeding 1 or whatever value there is no guarantee the overshot is detected. I have done this faulty in the Equilibre meters so I need to fix it with a code inbetween abs and M2F that sample and hold overshoots for a certain amount of time to be sure the overshot is detected and displayed.

trogluddite · 25 Oct 2011

I think it depends what you want from the max. I interpreted the question as - show me the maximum value that has ever happened since the last manual reset - in which case my little doo-dad should work fine when followed by an M2F to drive the display. I quite like latching 'max' indicators, as I'm often not in the same room as my VDU when setting levels - maybe that biased my interpretation. If building a peak reading meter, or 'flashing' overs indicators, then yes, you need something to make sure that they get shown for long enough to be seen - and then reset for the next interval.

tor · 25 Oct 2011

Ah.. yes, you are right. And thank you. Now I see how to improve the peak detector on my eq in a simpler and CPU efficient manner 🙂

wasabeevst · 25 Oct 2011

wow! Thank u all!!! Mo this osm is fantastic but I need to show the max value (like your scheme and the manual reset is cool) in dB... I want to show the same value like te meter of the chanel in our DAW (Cubase for example)... Pro Tools have a Peak Hold each channel in the display of the bottom... anybody know what i mean?

I want to show the maximum output value in dB... any idea?

infuzion · 25 Oct 2011

wasabeevst wrote: I want to show the maximum output value in dB... any idea?

Doesn't the SM's Toolbox have a compressor?

Mo · 25 Oct 2011

Here's a more precise Peak Hold Readout using trog's idea and coded in ASM. Readout is dB. Enjoy. Just click the display to reset.

trogluddite · 25 Oct 2011

Mo wrote: Here's a more precise Peak Hold Readout

Whoa, that boy is fast! 😀 Good work Mo. I can see a couple of tiny improvements, just to lower the CPU load a bit... 1) Tick 25 ought to be enough - the max value is held until reset, so it doesn't need a super-fast screen update. 2) 'Float changed' after the M2F, so that the maths and re-draw are only done at the times when the peak value changes.

Mo · 25 Oct 2011

trogluddite wrote: Float changed' after the M2F, so that the maths and re-draw are only done at the times when the peak value changes.

There seems to be no difference in the Trigger counter with a Float changed. I think it is already build into the M2F.

trogluddite · 26 Oct 2011

Mo wrote: Float changed. I think it is already build into the M2F.

Well, I'll be darned!! 😮 I had never noticed that before - I tried out a few little designs, and you're quite right, the M2F does indeed appear to have a built-in 'Float Changed! Very weird that it should be the only primitive with this feature (assuming it is the only one!) - but pretty useful, all the same. That's something else to add to my "exception to the rule" chapter of the trigger tutorial.

wasabeevst · 26 Oct 2011

wow thanks!!! that's what I need!!!!

CONGRATS! it's fantastic!

lesobrod · 29 Oct 2011

Hey people please let me up this thread. I like searching some weird and intricate processes so very need something like Min-Max detector. But it should do at least 2 detection - total, global Min-Max finding (until some Reset comes) and scanning for local Min-Max within some time period. Сomparison of all data helps to understand what is going on. I made a kind of module (and it works not bad) but it looks clumsy (I mean "made by axe"; is it correct word?). So maybe someone look inside and make some suggestion? 😉

trogluddite · 29 Oct 2011

Hi, This should be a bit of an improvement...

Seemed to me that the way that the displays showed the reset to zero, then rise/fall to the max/min was very confusing. In this version the min/max are calculated in Temp variables for the interval duration - then passed to the outputs once at the end - so the display now holds a steady min/max value for each interval.

Renaming min/max to minTemp/maxTemp also helps make the code easier to parse (for SM and for humans!) - it is not a good idea to have variables with the same name as a code command/function. I usually begin all variable names with a capital letter for this reason.

The ideal way for this to work would be so that the output shows the true min/max for the previous 'N' samples, constantly updated. I have wanted this for a long time, but the only way I find is to scan through a large circular buffer for every iteration, which IMHO is just too greedy for CPU. However, I think that the current method, used with an accurate stream timer, and something like Martin's pitch detector (to calculate the period) might be effective enough to get the local maxima for a good PSOLA algorithm. (I guess from questions in your other threads that maybe this is what you are looking for?).

lesobrod · 29 Oct 2011

OK this looks like made by laser! +100 to karma 😀 😀 P.S. Now I see that its allow to put many statements with ";" into one line of code. 😊 But how to do a hyphenation of long statement to the next line?

trogluddite · 29 Oct 2011

lesobrod wrote: But how to do a hyphenation of long statement to the next line?

LongStatement = ThisLine +AnotherLine +max(MoreLines,AsManyAsYouLike) +TerminateWithSemicolon;

That's it - line breaks are ignored by the parser, the statement simply ends with the semicolon - no need for any special 'hyphenation' characters or anything.