Falcon clock patch V4 series

[Badwolf]

I can't keep up with the discussion, I'm afraid. I've never had any noticible issue with clocks either before or after installing the patch.

But if delay/nodelay is an issue, I wonder about zero delay buffers? I've played with them before -- primarily when I was interested in clock doubling -- but could these be more useful in a clock patch than inverters and AND gates & whathaveyou?

AIUI they're PLLs that often offer multiple output drivers and can be configured to be multiples, divisors or exact replicas of the input timing.

BW

[exxos]

I can't keep up with the discussion, I'm afraid. I've never had any noticible issue with clocks either before or after installing the patch.

The problem only seems to show up with one Falcon with 2 different CT60s so far. The first I heard about this particular issue though. Though I cannot replicate the problem on my machine. The CT60 install says no buffers or inverters at all. So its expecting the combel clock directly otherwise (presumably) doesn't always work.

I also wonder if there was different speed grades grades of PLD used on the CT60. Maybe @Swissfalcon can see what PLD his has.

But if delay/nodelay is an issue, I wonder about zero delay buffers? I've played with them before -- primarily when I was interested in clock doubling -- but could these be more useful in a clock patch than inverters and AND gates & whathaveyou?

Do you have any particular numbers in mind ? The H5 uses zero delay buffers but I don't think they actually have a active push-pull output..

Though it may not be that simple because even just 33R can vary the timing by 2ns. my buffers only on the V4 only add about 3ns, but that plus 33R to solve the bad ringing.. your up to 5ns already.. Mix in other tolerances and it quickly adds up.

AIUI they're PLLs that often offer multiple output drivers and can be configured to be multiples, divisors or exact replicas of the input timing.

I think the ones internal in PFGA etc are OK, but I tried dedicated PLL chips way back in my early booster research and vowed never to use them ever again! Maybe technology has improved since then, but I don't really want to take the risk with them.

EDIT:

Just saw this which looks interesting...

https://www.infineon.com/dgdl/Infineon- ... 1cb50123a9

Bit expensive at around £10 a pop though.

[Badwolf]

But if delay/nodelay is an issue, I wonder about zero delay buffers? I've played with them before -- primarily when I was interested in clock doubling -- but could these be more useful in a clock patch than inverters and AND gates & whathaveyou?

Do you have any particular numbers in mind ? The H5 uses zero delay buffers but I don't think they actually have a active push-pull output..

The one I've had in my hands to play with is the ICS570A which only has the one output, admittedly, but it's 5V and can drive at 25mA.

Annoyingly though, the multi-output ones I can find on Mouser immediately are all 3V3 and only downto 10MHz.

Just an idea as I think this is exactly the sort of use case they're targetting, just a little slower and higher voltage than modern standards!

BW


EDIT: just had a massive bout of deja vu. I think we've had this exact discussion before down to the minimum frequency requirements and the ability to vary the lead/lag with capacitance on the feedback line.

[exxos]

The one I've had in my hands to play with is the ICS570A which only has the one output, admittedly, but it's 5V and can drive at 25mA.

Obsolete :(

Annoyingly though, the multi-output ones I can find on Mouser immediately are all 3V3 and only downto 10MHz.

The CY2308 is also 3.3V :( BUT, it can take 5V on it input clock. So what I could do is "fit" that before my current buffer chips to drive the normal 5V levels. The only slight problem would be feedback pin would need to be on the output of the buffers and that be 5V.. But I guess a simple as the voltage divider might be enough..

EDIT: just had a massive bout of deja vu. I think we've had this exact discussion before down to the minimum frequency requirements and the ability to vary the lead/lag with capacitance on the feedback line.

:shrug:

It might have been relating to the DFB1 at some point I guess.. but I think you had the V2 or V3 patch ? cannot remember, That uses inverters.. the V4 does not use inverters. I don't know if this would ultimately affect the DFB1, but at this point I assume it likely isn't as fussy as the CT60 is.

[mikro]

As for the resistors and CT60... as you might expect, this just opens a can of worms. I remember back in the day, a friend of mine had similar issues. R. Czuba instructed him to revert the clock patch (IIRC it was the Phantom) and put the resistors back. That helped to some extent but it had an unfortunate (but expected) side-effect: the sound crackles were back. So it's like alchemy... you add a resistor here, inverter there... and hope it will still work in sync.

(in the end, there was a new firmware revision for the CT60 and *that* fixed my friend's issues, with Phantom installed)

I'm curious about @Swissfalcon's results. I have one peculiar Falcon on the shelf, which a) works as is b) works with my old CT60 c) doesn't work with Centuriontech's CT63 (any Blitter access leads to an immediate freeze, only with this CT63 and only in this Falcon; I have a V2 clock patch installed there which first seemed to help but after a few days (???) it reverted back to its original behaviour).

So maybe the timing on the expansion slot really is quite critical.

[exxos]

.....
So maybe the timing on the expansion slot really is quite critical.

That is what I am thinking and why Czuba says remove the clock patches.. But then of course without proper buffering you get the SDMA issues back..I did complete the next V4 design earlier, but now I have abandoned that in favour for the V5 :roll:

The idea for the V5 is to have a zero delay buffer which has feedback to adjust to the combel timing accurately. It will still have the output drivers as on the V4, only the feedback will come from the buffer output. So whatever "load" is on the expansion board, will automatically adjust to the Combel clock. This way as far as anything plugged into the expansion port is concerned, like the CT60, it sees exactly what the Combel is putting out.

Secondly, the FPU/SDMA buffer will still be there and will still have some ns delay.. We don't really have a choice with that. So basically you get the best of both worlds.. In theory... But this will be rather a difficult project if only @Swissfalcon machine suffers from this particular problem.

I don't think there was any firmware updates from Czuba though ? I can't remember what mine had, whatever was flashed into my CT60e when I got it. Though I don't think Czuba did any changes after that anyway ?

The updated V2.01,I know I had issues with whatever that was, and went back to the original firmware V1.01c. This was on my original CT60. Not sure what my CT60e has off-hand.

[mikro]

I don't think there was any firmware updates from Czuba though ? I can't remember what mine had, whatever was flashed into my CT60e when I got it. Though I don't think Czuba did any changes after that anyway ?

Of course not. I'm talking about early days with 0.9x FW releases. :-)

[exxos]

I have pretty much done the V5 now. Though I won't really know exactly the resistor values as they are mostly trial and error. So I may have to do a second batch and the buffers are not cheap either :(. JLC do not stock the new buffer chips, so that will be a couple of weeks gone right there :(

While I can try it on my own CT60e, mine does not suffer the problems @Swissfalcon has. Ideally I would need to borrow his CT60e to see if it malfunctions in my own Falcon. But whole thing is a bit of a moot point currently as his CT60e has died and some tests need to be done first ideally anyway.

V5_ALPHA.PNG

[Swissfalcon]

I'm still trying some different things and measure them. I will post my result (can be tomorrow, busy day at work)

[exxos]

No worries, I'm not going to be around for the next couple of days anyway.

I have ordered all the parts for the V5 in preparation though.