I made a few discoveries while recording sounds from the CPC (through the stereo output, into a high-end sound card), and I'd like to confirm them thanks to the accuracy of an oscilloscope. Should you accept this mission, I would send you an executable on CPC to play some sounds, and if you could "record" what's going on on your oscilloscope, that would be great. I have no idea how you could conveniently show me the waves though. Maybe the stuff is now electronic and can produce nice waveforms at regular instants??? Forgive my lack of hardware knowledge, last time I used an oscilloscope was in high school :)
If these measures confirm my doubts, I'd be glad to share these info, which can be very useful for emulators and music software.
Thanks!
I can make a recording for you and send you the result in screenshots or data points which you can then analyse. You'll have to let me know what sample rate etc you need.
Bryce.
PM sent. Thanks a lot!
Yup, sent you some questions back. Many things can be measured on a scope, so your answers will make the results as close as possible to what you need.
Bryce.
Hi Targhan,
I can run some FFT tests if you want - these can tell you more in the frequency domain then oscilloscope screenshots,
John
I'm not sure he wants to go so deep as to investigate harmonics etc. But I can connect the CPC to a spectrum analyser too if needed.
Bryce.
Quote from: RetroCPC on 23:27, 12 April 21I can run some FFT tests if you want - these can tell you more in the frequency domain then oscilloscope screenshots,
Thanks, but indeed that's not going to be useful to me. Bryce's measures will be enough :). But I think I already know the result before even seeing it.
it will be useful for a proper replay
i did some "measures" back in time but it was with a soundcard sooooo i guess there was an hardware filter
i guess that square generation is all but squared :D
Quote from: roudoudou on 08:27, 13 April 21
i guess that square generation is all but squared :D
Oh man thats for sure :)
The CPC "Whine" from internal speaker is part due to Clock Phase Noise from the ULA + poor Grounding and PSU (both of which will result in poor phase noise as well as directly modulating the audio B/W).
Quote from: Bryce on 07:58, 13 April 21
...I can connect the CPC to a spectrum analyser...
The CPC is analysing the speccy, what will be the outcome? :laugh: :) :)
Quote from: GUNHED on 13:54, 13 April 21
The CPC is analysing the speccy, what will be the outcome? :laugh: :) :)
It's a different spectrum - Faster and no colour clash. :) (although it does have rubber buttons).
Bryce,
You need a sample rate higher than 48Khz?
I'm not a specialist but some project like MDFourier http://junkerhq.net/MDFourier/mdfourier.html (http://junkerhq.net/MDFourier/mdfourier.html)that work on emulator sound accuracy just need recordings of OG hardware with an external audio soundcard to make comparisons.
Can't wait to see some pictures of "what's really coming out" at the sound channel(s)
Quote from: stevensixkiller on 00:16, 15 April 21
You need a sample rate higher than 48Khz?
I'm not a specialist but some project like MDFourier http://junkerhq.net/MDFourier/mdfourier.html (http://junkerhq.net/MDFourier/mdfourier.html)that work on emulator sound accuracy just need recordings of OG hardware with an external audio soundcard to make comparisons.
I can sample at up to 2Ghz.
McBryce.
Quote from: GUNHED on 01:24, 15 April 21Can't wait to see some pictures of "what's really coming out" at the sound channel(s)
You can still have the results done by SuperSylvestre (http://cpc.sylvestre.org/technique/technique_son.html).
I just needed more tests, hence my post.
I'll try to get the tests done at the weekend. Unfortunately my "Retro-Desk" is currently my home office desk so I don't have the space during the week to set it up.
Bryce.
Quote from: Targhan on 10:52, 15 April 21
You can still have the results done by SuperSylvestre (http://cpc.sylvestre.org/technique/technique_son.html).
I just needed more tests, hence my post.
signals from AY chip or from a channel after amplification?
anyway, it's obvious 44KHz or 48KHz replay is not enough for CPC emulators
Did you allow Arkos tracker to use 192KHz replay when available?
Quote from: roudoudou on 12:07, 15 April 21Did you allow Arkos tracker to use 192KHz replay when available?
No filter on what frequencies to use is done, so yes. But I'm not sure you can make the difference, really :).
Quote from: Bryce on 12:04, 15 April 21I'll try to get the tests done at the weekend.
Thanks. I'm not pressuring you :).
Quote from: Targhan on 12:10, 15 April 21
No filter on what frequencies to use is done, so yes. But I'm not sure you can make the difference, really :) .
Thanks. I'm not pressuring you :) .
a SOUND 1,7 at 44/48Khz and 192KHz, the difference is obvious, the harmonics too :D
ok, that's not a usual sound
You don't even hear it on a real CPC either!
Quote from: Targhan on 12:19, 15 April 21
You don't even hear it on a real CPC either!
sound 1,7 is still clear for ME on a real CPC
at my age sound 1,5 is too high-pitched but i remember when i was young, up to SOUND 1,3 :D
are you still ok? => https://www.youtube.com/watch?v=qNf9nzvnd1k
On the CPC speaker? I *think* I hear something at 1,5, but there's so many glitch it could be anything. Even a FOR loop in Basic produces sound, so what you hear with the SOUND command is really biased.
Quote from: roudoudou on 12:21, 15 April 21
sound 1,7 is still clear for ME on a real CPC
Same here, for sound 1,6 I do need external speakers though.
Quote from: Targhan on 10:51, 10 April 21If these measures confirm my doubts, I'd be glad to share these info, which can be very useful for emulators and music software.
Very interesting topic! Looking forward to what you will be sharing.
PS: SOUND 2,7 here (it's easier in stereo)
Why does the program crash the CPC and mess up the screen? Is that correct? Surely you could have just created those sounds in BASIC?
Bryce.
Edit: Ok, I'm going to need a better program to work with. I need each tone in a separate file and it should repeat when I press the space bar or some other key (or be continuous. Otherwise this test could take days to complete.
>Why does the program crash the CPC and mess up the screen? Is that correct? Surely you could have just created those sounds in BASIC?
Huh, there is a basic loader loading and executing a binary program. I tested all this on a real CPC, it works fine. I created them using AT2 because it gives me more control over the sounds I want.
Please PM if you want something else, you have my email.
Edit: ok it seems you want each sound in a different program. I'll do that and sent it again to you. I don't know your process, I expected you to record everything, then "simply" take "pictures" of the signal at precise instants.
No, that's not the way it works. You can take long samples, but you loose the resolution. To properly zoom in on a single wave and see the details, you need to be taking lots of samples over a short time.
Bryce.
Ok no problem. I'll come with another program :).
Done, PMed you.
14kHz is the upper limit for me. :'(
Hi,
I had some time this morning and set up the test. Unfortunately, most of the sounds are non-symetrical, so they don't trigger properly and don't really tell much.
As the frequencies are so low, I can't measure them on a Spectrum Analyser. They only start at 9kHz.
Bryce.
Thanks a lot for this!
- If the period is too low, and if you have time and want to make another test, simply change the "v=&ef" value line 1000 in the code I sent you. The smaller, the higher the sound.
- I guess the yellow line is the left signal, the blue one the right?
- From what I see, the signal seems perfectly square. No "round curves". Can I deduce from that that the sound generation of the PSG is pretty accurate and does not produce "lazy" square envelopes?
The squareness of the wave is also due to the very low frequency. The higher it goes, the rounder it would become.
When I get time I'll do some examples of higher frequencies.
Bryce.
That would be great, thanks!
The funny thing is that, from your standpoint, you consider the frequency as "very low", but they are actually pretty high, musically speaking.
Quote from: Targhan on 11:56, 27 April 21
That would be great, thanks!
The funny thing is that, from your standpoint, you consider the frequency as "very low", but they are actually pretty high, musically speaking.
The base frequency of your sample is approx. 262Hz which is middle C on a piano, so it's dead centre from a music perspective, but on the scale of what a human can hear (20Hz - 20kHz) it's pretty low down on the scale* and from a physics / electronics perspective it's extremely low.
Bryce.
*Yes I'm aware that hearing sensitivity is logarithmic so it's not quite as low as I state.
Quote from: Bryce on 08:19, 27 April 21The squareness of the wave is also due to the very low frequency. The higher it goes, the rounder it would become.
Which is usually a sign of an oscilloscope reaching its bandwidth / sample frequency limits, right? For the CPC clock, this shows up even with decent enough starter oscilloscopes. E.g., with my 80 MHz Hantek, the CPC clock doesn't look squareish at all, but already much more like Sinus Wave.
I guess 100 Mhz is already the min for 4 MHz system. Well, square waves have high frequency harmonic content in its Fourier spectrum, and if you can't capture these due to bandwith limits of the scope, it gets roundish.
For audio, it shouldn't be a problem though. Not even for these B***S*** 50 $ pocket oscilloscopes on EBay.
The noise looks so cool!
Also, there are pretty fast state switches. High frequency by the PSG, and 3 octaves lower square modulation by hand?
Were you tickled by BSC's great gritty sounds?
You're going to be disappointed, it's only basic square sound, without then with noise, then noise only :).
Oh, so in sound4.png, you have a sound in each channel? Even if it's the case, that's doesn't square for me!
Yes, your father and I are very disappointed by your behaviour!
Quote from: m_dr_m on 13:56, 28 April 21Oh, so in sound4.png, you have a sound in each channel? Even if it's the case, that's doesn't square for me!
The sound is only played on the center channel, so what I see on sound4 is the same signal L and R. It looks very square to me.
That's because it is square. Sound signals don't need to be a sinewave.
Bryce.
Sound1 is square.
In Sound4, there is state alternance inside the slower state alternance! (not taking about the noise (*) but the tiny fragments on either high or low state).
That is typical of 'ring square modulation' (multiplication of 2 square signals).
It could also be achieved by a very fast and well synchronised arpeggio (i.e. the switch of periods is done at same frequency than 2* the lower sound).
If you can generate either from basic, you have all my attention.
(*) Well, maybe it's just the pseudo-random state of the noise staying constant for a while.
Quote from: VintageAdvantage on 16:05, 27 April 21
Which is usually a sign of an oscilloscope reaching its bandwidth / sample frequency limits, right? For the CPC clock, this shows up even with decent enough starter oscilloscopes. E.g., with my 80 MHz Hantek, the CPC clock doesn't look squareish at all, but already much more like Sinus Wave.
I guess 100 Mhz is already the min for 4 MHz system. Well, square waves have high frequency harmonic content in its Fourier spectrum, and if you can't capture these due to bandwith limits of the scope, it gets roundish.
For audio, it shouldn't be a problem though. Not even for these B***S*** 50 $ pocket oscilloscopes on EBay.
It can be caused by the scope not having the bandwidth, but in the case of a sound signal obviously it would be other factors causing the roundness. Parasitic capacitance and inductance will round the wave, transistor switching times will round the wave and the miller effect can also cause it.
For a good looking square wave you need a scope with at least 10 times the bandwidth of the fundamental wave so you would expect a 40MHz scope to be enough for 4Mhz. However, the sample rate is also important, so even a 100Mhz scope will show a strange square wave if it the sample rate is too low.
Bryce.
Quote from: m_dr_m on 14:47, 28 April 21In Sound4, there is state alternance inside the slower state alternance!
You're right. No idea what they are! Like I told you, it's only sound + noise, generated in Basic...
So, Targhan, can you already conclude what you were trying to find out here?
Quote from: BSC on 12:33, 30 April 21
So, Targhan, can you already conclude what you were trying to find out here?
I don't think
I am the one would can conclude, only based on screenshots. Bryce can, but maybe we should wait for new recording using higher frequencies.
Basically, I just wanted to be sure that the PSG was producing perfectly or almost perfectly squared envelopes (as Sylvestre showed in his page which I linked earlier). When recording music using a sound card, I noticed very strange stuff, which I thought was because of the PSG. It
seems to turn out it is just because of filters modern sound card have.
So basically, it seems the PSG doesn't do crazy stuff :). I emulated some of the strange stuff I noticed, in Arkos Tracker 1 (mostly when using noise), but removed it for AT2 (which turned out to be a wise move).
This might go above the heads of many here, but just for the fun of it, I analysed the highest frequency the CPC can create: Sound 2,1 which is for me at least not possible to hear.
The higher the frequency, the more expected distortion (and the easier it is for me to analyse the spectrum).
The first screenshot is taken on a 100MHz oscilloscope. This means that I am theoretically measuring up to the 1611th Harmonic (if it actually existed), so any rounding seen is due to the limits of the CPC and not the scope. As you can see, the wave is still relatively square. The rounding on the leading edges will be due to capacitance on the CPC's PCB and parasitic properties of the resistors used to merge the channels. The noise seen on the top and bottom of the waves is most likely the CPC's clocks and switching noise from other parts of the circuit. But it's at a frequency that no speaker could reproduce, no ear could hear and it's volume is tiny anyway (20mV peak to peak).
The second screenshot is a spectrum analysis of the same signal. Here you can see the fundamental frequency on the left (62.19kHz) plus the 3rd, 5th, 7th and 9th harmonic. There is a tiny 4th, 6th and 8th harmonic which contributes to the rounding, but as you can see they are 30dB below the fundamental frequency and would normally be ignored in most calculations unless I was dealing with highly sensitive GHz signals.
Btw, the actual frequency is 62,190kHz as the SA reports, not the 62.5kHz that the Oscilloscope reports. Spectrum analysers are much more accurate than scopes for measuring frequencies. The ramp up of the noise floor on the left is because the SA is approaching it's lowest frequency limit (9kHz).
So to answer the original question: The CPC produces an extremely clean, as square as possible (for the electronics in a CPC) wave.
Bryce.
Bryce,
Is the component on Marker 1 the square-wave frequency?
If so, the SNR is very poor - with a wide band Spectrum NF only 30dB down?
When you short the input of your SA do you still see such a rising noise-floor?
Listening to the CPC speaker "whine" I would expect the NF to be rising as seen on your SA - in fact worst if the SA lower limit is 9KHz.
You can see the fuzz on the Square wave peaks gives you an idea of the wide-band noise levels.
As you say, to be expected from a consumer "budget" home computer design back in the 80's.
Marker 1 is the fundamental frequency, but the SNR is not a component of the CPC, rather the fact that I am measuring a 62kHz signal on a Spectrum Analyser that has a range from 9kHz to 3.2GHz. So as expected the noise floor is raised at both ends of that spectrum and the signal I'm measuring is at the very lower limit of that spectrum. If I short the SA input, obviously, the noise floor drops to the level of the SA'S internal noise and is almost a straight line.
The CPC is an 8 bit computer, why would I expect the SNR to be anywhere close to some high-end audio equipment? The "fuzz" as you call it is the switching noise from the CPC's circuitry and at 30dB down is pretty good for a non-shielded TTL circuit with a cheap-ass AY where they didn't even bother to have separate grounds for the analogue and digital portions.
Bryce.
Hi Bryce,
I'm not convinced that the noise at the bottom end of the band is due to the lower limit of your SA. Sure the close in Phase noise of an "affordable" SA local oscillator is not going to be anything to write home about, but I doubt its as bad as seen in the sweep - I think much of what your seeing is really the SNR of the CPC.
For sure the HF noise is due to the lack of Ground plane, decent PSU decoupling etc. If I had spare time on my hands I'd love to redesign the PCB for improved video / audio :) With modern Hi-bandwidth LCD monitors you can see a lot of noise on the CPC's RGB output - even with the HF BW limiting filter capacitors fitted...
Its not in anyway a criticism of the CPC :) , its just the reality of 30 years of engineering progress in consumer grade electronics...
If I get time later I'll simulate the same frequency on an expensive signal generator and compare the noise floor in that range.
Bryce.
I don't understand half of what you're talking about :). But the conclusion seems to be straightforward:
Quote from: Bryce on 14:44, 01 May 21The CPC produces an extremely clean, as square as possible (for the electronics in a CPC) wave
You hear that @roudoudou (https://www.cpcwiki.eu/forum/index.php?action=profile;u=1714) ? Like SuperSylvestre had hinted with his measurements.
Thanks a lot Bryce!
Quote from: RetroCPC on 19:52, 01 May 21
Hi Bryce,
I'm not convinced that the noise at the bottom end of the band is due to the lower limit of your SA. Sure the close in Phase noise of an "affordable" SA local oscillator is not going to be anything to write home about, but I doubt its as bad as seen in the sweep - I think much of what your seeing is really the SNR of the CPC.
For sure the HF noise is due to the lack of Ground plane, decent PSU decoupling etc. If I had spare time on my hands I'd love to redesign the PCB for improved video / audio :) With modern Hi-bandwidth LCD monitors you can see a lot of noise on the CPC's RGB output - even with the HF BW limiting filter capacitors fitted...
Its not in anyway a criticism of the CPC :) , its just the reality of 30 years of engineering progress in consumer grade electronics...
Ok, here's a comparison. Same frequency and amplitude, but produced by a high-end signal generator. As you can see the even harmonics are gone, but the rise in noise floor at the lower end of the scale is still there (ie: it's a component of the SA's filtering). The slightly higher peaks are due to the fact that I have to use a different cable to connect the equipment to each other. On the oscilloscope screenshot the differences are much more obvious, the wave is perfectly square with no rounding on the leading edges. This is due to the fact that the wave is being produced by high end equipment that can produce signals up to >100MHz and has high frequency output transistors.
Bryce.
@Bryce (https://www.cpcwiki.eu/forum/index.php?action=profile;u=225) when and if you have time, could you record a sound with a period of 0? I know the Basic allows that but I hope it won't implicitly raise it to 1. It should produce the same period as 1 (like what you previously recorded), but according to Zik experiments some years ago, the volume should be lower because of a "filter" (please don't ask me for the details, I don't know them) that is produced by the PSG with this very period. It would be interesting to know the reason...
Bryce,
Thanks for the extra tests - I'm really surprised that the Close-in LO phase noise of the Rigol is so poor - I wonder what they are using as the LO synthesizer?
What CPC machine are you using & are you Measuring via the 3.5mm audio output?
I have a CPC6128, you now have me interested to run a few measurements, whats the basic commands to generate a 1KHz / full level signwave via the CPC sound chip?
@Targhan (https://www.cpcwiki.eu/forum/index.php?action=profile;u=110) : Sound 2,0 doesn't produce any signal, the output stays a 0V.
@RetroCPC (https://www.cpcwiki.eu/forum/index.php?action=profile;u=3752) : I have used much more expensive analysers at work (Agilent E and N series) and they all do this. It is not due to the local oscillator, rather the fact that the closer you get to DC, the harder it is to filter noise without attenuating the signal too much. That's why almost all SA's start at 9kHz. Usually this doesn't matter though, because they aren't really designed to analyse single signals this low down.
By the way. I should mention that the RF Pre-Amp was turned off for these measurements. With the Pre-Amp turned on the overall noise floor drops a few dB more and the low frequency noise floor is flattened slightly more (see below).
I used a standard German 6128. Measurements were taken at the 3.5mm output. To make a sound you use the command Sound 2,x,y,z
2 is the channel
x is the tone (1kHz is around 61 or 62)
y is the length of time it plays for (30000 gives you about 5 minutes)
z is the volume (I think 15 is full volume)
Bryce.
According to the 464 User Manual, tone period is 125000÷frequency, so 1kHz should be a tone period value of 125.
Bryce,
Thanks for the sound commands - Ill give it a go.
Spectrum analyses sweep the frequency band by mixing there internal Local oscillator with the input signal - the sloping noise floor that trends upwards towards DC is the result of the close-in carrier phase noise of the SA's local oscillator (and to a smaller degree any active internal gain stages).
Its true that all oscillators exhibit increasing close in phase noise towards DC – but a good design will reach the flat portion of its phase noise noise floor before say 1KHz.. or as quickly as possible – the higher the Q of the oscillator the narrower the oscillators "side skirts". Tunable oscillators by there very nature have lower Q hence worst close-in phase noise.
https://www.anritsu.com/en-US/test-measurement/solutions/ms2840a-066/index
Its explained very "simplistically" in the above link - see Figures 1 and 2... where the (marketing departments) curde graphs give "text book" examples of LO close-in phase noise...
WRT SA filters (RBW & VBW) in older units they where "simple" Analogue filters which placed practical limits on the minimum resolution B/W - the brillent HP3585A/B has a 3Hz B/W for its narrowest filters - but modern SA supplement wider B/W analogue filters with FFT digital filter techniques to achieve sub Hz resolution bandwidths with comparative ease.
I had RBW and VBW set to 100hz, otherwise the scan takes forever. Of course it's unfair to compare the Rigol with an Agilent or Anritsu. My Rigol "only" cost around €5000, whereas the price of the Agilents we use at work is on the wrong side of €70,000 !!
Bryce.
Amstrad CPC6128 Audio Line output 1KHz test tone:-
Sound 2,62,xxxx,15
0dB = 137mV RMS into 20Kohms
16K FFT
Well the Amstrad Sound chip produces Squarewaves - rather then "clean" low THD Signwaves
Lots of LF noise, but could be worst...
Quote from: RetroCPC on 19:29, 02 May 21
Lots of LF noise, but could be worst...
Only the internal speaker, not the 3.5 mm jack. :)
Judging by the whinny noise coming from my CPC6128 speaker, the internal audio circuit must be much worst, I'll repeat the test measuring across the internal speaker...
Quote from: GUNHED on 19:32, 02 May 21
Only the internal speaker, not the 3.5 mm jack. :)
The above FFT is measured from the line output...
Your ears work better. :laugh: (Didn't see the picture before)
Well apart from the interesting noise floor, I had no idea that the CPC sound chip produced Square waves rather then Signwaves...
I'm sure the capacitance and inductance of the end stage will ensure that they are anything but square :D
Bryce.