The Famicom's and the NES's Audio Path

Apologies for resurrecting an old thread, but it seemed the most appropriate place to ask my question...

What signal / voltage level is expected at that 22K resistor, coming in from the NES expansion port?

I would like to directly attach a 0-5V PWM signal from a microcontroller, but my feeling is that is going to overload or distort the input, and will probably need some extra circuitry in place?

OK, some more digging, there aren't 3 x 22K resistors - actually 2 x 20K and 1 x 12K.

Looking at schematics, 'Audio In' (pin 3 on the Expansion Port) feeds into R9 (20K resistor) before the 'amplifier' circuit... so definitely seems like putting 5V into a circuit that expects something much smaller, needing amplification, is a bad idea.

Would adding in a 47K resistor for a total of 67K resistance (as seems to be the standard for audio coming from a PowerPak) be enough to safely input the 5V signal?

It's hard to give you a straight answer without directly measuring the voltages on CPU pins 1/2.

You can be confident, however, that if you choose the same voltage range as CPU pin 1, due to the 20kΩ resistor from the expansion port, it'll be the same voltage (loudness) after all three signals are mixed together.

If you want to use a 5Vpp source, and CPU pin 1 is less than 5 Vpp, then you should be able to effectively match the volume by setting the currents to match—e.g. if Vpp(one pulse wave at maximum volume) is 1V, and you want your output to be the same volume, then adding 80k in series should make the volumes match. (80k+20k=100k ; 100k/20k = 5 = 5V/1V)

Absolutely. Your reply was very helpful though, thanks!

Dug out my PropScope and the NES today and took some measurements from those CPU pins. Showed a voltage range of around 0.3V (ranging from around -0.226V to +68.9mV).

So I should be aiming to attenuate that 5V signal down to around 0.3V then?



PS. Testing was done using the 'test2-mmc3.nes' ROM from https://ahefner.livejournal.com/12368.html - pulse set to full volume:

Depends on how many different audio signals your microcontroller is generating.

If you're just generating a single pulse wave, and you want that pulse wave to be the same volume at maximum as the NES's pulse waves, which are 0.3V, then yes; this attenuation could be achieved by putting an extra ≈313kΩ in series with R9. On the other hand, if you're mixing 4 waves, and you want all of them to be the same volume, you'd want to add ≈63kΩ.

But ... I have to admit I'm a little surprised it's only 0.3V. I'm also really surprised the voltage went negative: that's supposed to be a unipolar pull-up DAC inside the CPU (with an external 100Ω pulldown resistor)—nothing should be creating negative voltages here.

I read something about negative current here:

http://nesdev.com/2A03%20technical%20reference.txt

+-----------+
| 4-bit DAC |
+-----------+
"This is just a standard 4-bit DAC with 16 steps of output voltage
resolution, and is used by all 4 sound channels. On the 2A03, rectangle wave
1 & 2 are mixed together, and are available via pin 1. Triangle & noise are
available on pin 2.

These analog outputs require a negative current source, to attain linear
symmetry on the various output voltage levels generated by the channel(s)
(moreover, to get the sound to be audible). Instead of current sources, the
NES uses external 100 ohm pull-down resistors."

Was reading another thread earlier that may help indirectly explain too, that this negative DC offset out of the CPU pins is corrected in the mixing / amplification circuit:

viewtopic.php?f=3&t=11432#p131160

Would be great if somebody else could confirm my readings on their own NES though.

I have an oscilloscope, but unfortunately nothing resembling a useful test cart. (I still just have the ridiculous m218 cart I made)

Taking a random game and looking at pin 1, I see voltages approximately around 0.3V(max) and 0V(min)... so I guess the range is accurate.

(Did you set your 'scope to AC coupled? Or failed to connect ground? Not important, tho.)

Had to look up Mapper 218 - love it!

Good to at least have confirmation of the 0.3V - thanks again for all of your input on this!

I have an assembler. What would you like to see in a mixing test cart designed for mapper 218?

If you're offering your time...

There's two separate measurements to be made. Pin 1, for the pulse waves, is simpler. There doesn't seem to be any realistic bandwidth constraints, so just picking the maximum frequency, a 3/4 duty cycle, and letting me adjust the volume for each of the two channels independently, would be good enough.

Pin 2 is a little hairier because the triangle wave can't be set to a fixed value — at least not without trying tying CPU pin 30 to CPU A3. It doesn't seem wrong to just ignore it and just set the pulse wave volume and DPCM DAC. Although maybe some cleverness can ease measurements of the triangle wave as well?

... huh, to describe the actual problem is to put it in scope.

You can advance its position by a fixed amount, though. If you're willing to take 32 measurements you could put it into every phase? (Or with live/interactive measurement you really could just set the phase how you want this way.)

I forget: Does the Reset button do anything to triangle phase? Or should this be included in the test ROM?

The triangle phase, tri_c0 through tri_c4, is cleared on reset. For example, check visual2a03 t13993.

Did some testing of my own because I'm looking to do the same thing!

My test is pretty simple. I've attached it in case anyone else want to try. The rom only runs from $E000-FFFF so it should be fairly mapper independent assuming the last bank is fixed. Visually it doesn't display anything worthwhile, just some junk from my CICOp register test.

Only SQ1 & SQ2 channels are enabled, SQ1 is set to C4 (middle C) and SQ2 is set to E4. Both have 50% duty and max volume.

I did not find any negative voltage. I found a max output level of 350mV (note 10x probe used in scope shot) this is when both channels are active (summed). When only one channel was pulsing high, it's output is 200mV.

I can run some other simple tests like this if it's of interest, but this pretty much answered my questions.

Great minds

...and you plan on adding 313K of resistance to compensate? or you have a better solution in mind?

What's your thinking on 313K? 0-5v signal from mcu, trying to get down to 200mV is a factor of 25. Thus you need 20k * (25 - 1) = 480K..?

Being a PWM DAC, it's fairly easy to reduce the synthesized channel to less than 5v. In an 8bit DAC a simple implementation would be for a square channel to have 4 bits of volume similar to the APU/VRC6/etc. So the channels max would be a DAC level of 15 out of 255 equates to 5.9% of 5v = 295mV for that channel without tweaking anything. A little beyond the target 150-200mV per channel to match the APU.

Couple ways to tweak things to match. An added 20K resistor would work, but I kinda like/have to assume many people simply soldered in a 47k for the audio mod. That's my preferred condition, and then adjust the PWM levels to match. So 67k divides that channel by 3.35, so 295mV -> 88mV equavalence if I'm understanding everything.

To get to a volume comparable to the APU, the level needs doubled bringing it to ~175mV. That can be done in the mcu by shifting the final output volume left by 1 bit. Sort of cuts an 8bit DAC down to 7bits in terms of number of total full volume channels prior to clipping, but that's manageable. 7bits would provide room for quite a few full volume channels at once. The lower bit really isn't lost either, channels could simply have 5 bits of volume to equal an APU square if they've got a use for the extra resolution.

P.S. I'm guessing I've made an error in my math/knowledge somewhere..

Was just going on lidnariq's numbers from earlier in this thread:

viewtopic.php?f=9&t=56&start#p205492

Both CPU pin 1 and Expansion Port pin 3 feed into separate 20K resistors in the audio mixing / amplification circuit, so to match CPU pin 1's 0.3V as I understand it: 5V / 0.3V = 16.666666667 * 20 = 333.333333333 minus the 20K already in place equals 313K?

My problem is that I would not be using a DAC with amplitude settings, and be able to just reduce the signal down that way. I'm looking at a PWM pin on an 8-bit AVR, or likely just software PWM on a digital pin, so it's either on or off, 0 or 5V, full stop.

It may actually be a benefit if an ext audio synth is a bit quieter than the internal APU.

This is because the steps between the 15 audio levels are quite coarse. Less full volume = finer steps, leading to smoother envelopes; something the APU can't* offer but a synth with a similar 4-bit interface can. That would enable using slow-moving envelopes (in a satisfactory/transparent way) more readily.

The max volume of the APU squares aren't very useful, IMO. There's many times i've wished for an "x.5" setting but i've never felt the need for maxing the squares. Rather, i avoid it.


*There's a workaround for the apu wich is chorusing two square channels and then you can kind of get doubly fine volume steps at the expense of independent movement/having the sound to be slightly detuned. Sometimes, you can even hide the coarse stepping in a second square voice at another interval than unison, but it is circumstantial (dependent on the harmonics of the tone, i presume).

313K assumes the APU channels have a max volume of 300mV. Take it for what it's worth, but based on my testing a single APU square @ max volume is more like 150-200mV. 300-350mV is both squares summed, I'm guessing that is what your and lidnariq's tests were measuring.

Oh my bad, I read too far into your statement, I thought you were implementing a PWM DAC too. Your single square channel with invariable volume will be pretty limited.. Why not add a cap to your resistor to unlock your 8bit AVR's PWM DAC ability.? Give yourself ability to adjust the channel's volume, even allows multiple channels of varying shape if you want to get fancy. Obviously up to you, but just wanted to make sure you were aware of how simple a PWM DAC is to implement on a 8bit mcu. Open music labs has a pretty good article that showed me the light, along with an arduino sketch tutorial.

As always, thanks for the insight, these are the types of things I don't have a good sense for.

Exactly.

If the OEM pulse channels are 150mV each into 20kΩ, then that's 7.5µA at max; 5V and 7.5µA is 670kΩ. To match the NES's built-in 20kΩ and the typical 47kΩ added, I'd assume you'd want a ≈600kΩ resistor out of the PWM DAC ... if it's still only one channel being mixed in.

Personally, I'd be tempted to mix multiple channels.

It's completely ridiculous, but you could add an external resistor DAC to add volume control without having to manage the IRQs from a PWM DAC.

Something like:


And you could then toggle bits in DDRA to adjust the volume.


I say "ridiculous" because in practice, your microcontroller is probably not too busy to manage updating the PWM value to serve as a DAC.


In thinking about this, it occurs to me that we could sink current out of the mixing node to add software-controlled master volume control.

If we had something like this:
although it might be more useful to pull the mixing node up to 5V (instead of pulling it down to ground as in this schematic) for greater dynamic range.

Sounds like it could be useful for:
-fade in/out
-half-mute (for example in a menu).
-tri envelope when playing solo
-super-exponential decays/swells of squares/noise/samples playing solo or with shared rythm
-maybe even improved dynamic range and/or fizz reduction of samples



For one thing, many studio synthesizers use a swing of 5v or even greater for audio prior to the output stage - this to keep the noise floor down relative to the signal on analog audio lines. This becomes increasingly important when mixing a large number of signals; effectively raising the noise floor.

Consumer-grade outputs however have a nominal swing of 0.447V peak-to-peak using a single sine, or more importantly 0.316V root mean square (pretty close to the NES?)

Studio equipment should have a nominal output of 1.228Vrms.

I think Sunsoft 5B audio breaks the consumer-grade nominal levels by quite a bit. It has one benefit, though:
using its coarse, loud squares as the reference point, it'll relatively push the full APU dynamic range into useful range and make its steps seem finer by comparison. Use Sunsoft for quick and loud envelopes, and the APU's for "quiet" and more nuanced/slow ones. This has the unfortunate side effect of the hypothetical game/music becoming the loudest in the nes library and will see users reaching for the volume control of their tv/speakers/laptop, but fwiw it almost works like having a finer envelope range.

Then the sunsoft synth has that 5-bit internal hardware envelope to mitigate its coarseness, but no game has ever used it...

This ability would be significantly more limited if operating with the assumption of a fixed 47k EXP port audio mod right? Gives more reason to include the pot in my EXP port dongle..

I'm a little confused how one would integrate expansion audio channels and master volume at the same time in terms of the PWM DAC's output level calculation.

Rather than BJT tomfoolery, it's probably better to just pretend we're going to use a digitally controlled potentiometer. Then it's a lot easier to see what's going on:



Quickly playing around with falstad's circuit simulator, assuming pin1 is a 150mV(max) square wave and pin2 is just 100Ω to ground, adjusting the pot between 0 and ∞ yields 41mVpp to 56mVpp ... that's not very good, only a couple dB

That 20k resistor is a pain.

Fine, so we add a negative impedance converter to undo the 20k that we can't get rid of. It'll require that we synthesize a negative voltage, but this already crazy analogland, so whatever.

And, yup: that lets us get an arbitrary volume attenuation. Synthesizing a different regulated choice of "ground" for the NIC other than 0V should make it possible to not need sophisticated power circuitry to create a negative voltage:

A 20k pot here would become a -20k pot. When set to -20k, we've magically made a 0 ohm resistor and attenuation is ≈infinite. When set to 0, it's only 3dB (20·log₁₀(56÷41)) quieter than without this add-on hardware.


The math becomes gross but it should still be doable. An electrically super simple and lazy option would be to replace Vreg with the external mixed in audio signal, but the mixing equation becomes a little weird.

—

If we're going to be even more completely ridiculous, a NIC allows us to multiply the complex reactances of three different passives. By changing some of them to have some imaginary component (e.g. a capacitor or inductor) we can instead have a programmatically controlled high-/low- pass audio filter.