Shop OBEX P1 Docs P2 Docs Learn Events
SA tracking generator as field-expedient P2 clock source? — Parallax Forums

SA tracking generator as field-expedient P2 clock source?

JRoarkJRoark Posts: 1,215
edited 2019-12-05 04:24 in Propeller 2
I want to go hunt the upper limits of the P2’s non-PLL speed capability, but I lack a suitable clock source. Tonight it hit me that I might be able to use the tracking generator from my spectrum analyzer for this task. The output does -20db to 0db, which from my understanding, if capacitively coupled to XIN, should be in the normal crystal range.

Am I milking without a bucket here, or would this be workable? I’d really want to see how far the P2 can be pushed with the PLL out of the way.
«1

Comments

  • evanhevanh Posts: 15,916
    You'll need to chill the prop2. And the faster you go the more powerful the chiller has to be.

    The logic stability threshold is on a slightly different thermal slope to the PLL natural limit. At higher temps, say >20 oC, the PLL runs slow enough to prevent crashes from logic timing limits. But at lower die temperatures the PLL gets fast enough to cause crashes if you don't cap the set frequency. The exact limit of that cap is higher as the die temperature is lowered.

    This then leads to the need for the thermal pad to be placed directly on a real heatsink rather than a bunch of vias in the PCB.

  • evanhevanh Posts: 15,916
    So, a point there is, you don't need a programmable external clock source to get a nice speed up. But you do need a chiller and remount the prop2 suitably.

    You won't get far at all without that.

  • 0dBm is .2236 volts RMS into 50 ohms. That's .632 volts peak to peak. You might get more without 50 ohm termination. The P2 oscillator results in 1.4Vpp at XIN. I'm not sure what the P2 really needs. Maybe it would work at normal clock frequencies, but >300MHz could be a problem. I would say, "try it," but start with the standard 20MHz to check.

    FWIW, many RF mixers need more than 0dBm input power for their local oscillator. 10dBm is 2Vpp into 50 ohms. It's easy to find a wideband amplifier with 10dBm output.
  • I have a $5 clock gen option for my next board that goes up to 500MHz. I am doing another revision to allow for some cheaper microSD and microUSB connectors so I will have this clock gen including a trimable 0.5ppm oscillator.
  • jmgjmg Posts: 15,173
    JRoark wrote: »
    I want to go hunt the upper limits of the P2’s non-PLL speed capability, but I lack a suitable clock source. Tonight it hit me that I might be able to use the tracking generator from my spectrum analyzer for this task. The output does -20db to 0db, which from my understanding, if capacitively coupled to XIN, should be in the normal crystal range.

    Am I milking without a bucket here, or would this be workable? I’d really want to see how far the P2 can be pushed with the PLL out of the way.

    It's going to be tough to wiggle a pin at CMOS levels to 300MHz+, as well as get it past the P2 pin amplifiers.
    Most clock generators only spec 250MHz on their CMOS modes, but maybe you can 'overclock' them.
  • cgraceycgracey Posts: 14,155
    I think it should be no problem to drive 400MHz into XI, if you have it. Just select mode %01_10 via HUBSET.
  • Great input all. Thank you!

    I ordered a cheap 10 db stripline-type broadband amp. This will hopefully arrive tomorrow and I’ll have some data to share over the weekend.

    Chip’s comments intrigue me. A P2 at 400+ mhz would be amazing if it was proven stable. And since Peter is prepping the P2D2 for exactly this possibility, it makes warp speed very accessible to us mere mortals without much hacking.

    This is gonna be fun.
  • cgraceycgracey Posts: 14,155
    edited 2019-12-06 06:32
    If we could use a 14nm FinFET process, we could probably get up to 8GHz and overclock even faster.

    A 3-inverter ring oscillator in our 180nm process runs at 5GHz.

    A 3-inverter ring oscillator in the 14nm process run at 200GHz. That's 40x faster.

    Who knows what would be possible?
  • Cluso99Cluso99 Posts: 18,069
    Wonder what a P1 on 180nm would run at :wink:
  • evanhevanh Posts: 15,916
    What sort of clock rates did the FPGA versions of the prop1 get to compared to prop2 FPGA rates?
  • cgraceycgracey Posts: 14,155
    edited 2019-12-06 09:32
    evanh wrote: »
    What sort of clock rates did the FPGA versions of the prop1 get to compared to prop2 FPGA rates?

    I think they both ran around 80MHz, but we used much newer FPGA's to develop the P2, which is a much more complex design than P1.
  • evanhevanh Posts: 15,916
    II thought Cluso and others had a modified prop1 running on newer FPGAs too. More addressable hubRAM, port B and what not.

  • cgracey wrote: »
    If we could use a 14nm FinFET process, we could probably get up to 8GHz and overclock even faster.

    A 3-inverter ring oscillator in our 180nm process runs at 5GHz.

    A 3-inverter ring oscillator in the 14nm process run at 200GHz. That's 40x faster.

    Who knows what would be possible?

    A Prop with a Ghz speed rating does set the mind aglow with possibilities.

    Toto... We’re not in Kansas anymore!

  • cgraceycgracey Posts: 14,155
    edited 2019-12-06 16:15
    JRoark wrote: »
    cgracey wrote: »
    If we could use a 14nm FinFET process, we could probably get up to 8GHz and overclock even faster.

    A 3-inverter ring oscillator in our 180nm process runs at 5GHz.

    A 3-inverter ring oscillator in the 14nm process run at 200GHz. That's 40x faster.

    Who knows what would be possible?

    A Prop with a Ghz speed rating does set the mind aglow with possibilities.

    Toto... We’re not in Kansas anymore!

    4k HDMI wouldn't require any special hardware, just (fast) pins.

    In 14nm technology you can get 2MB (2M bytes, not bits) of hub RAM in 1 square mm. 16MB would be no problem.

    At 8.6 GHz, a 32-bit counter would roll over TWICE every second.
  • BeanBean Posts: 8,129
    Stop teasing us Chip.
    :wink:

    Bean
  • JRoarkJRoark Posts: 1,215
    edited 2019-12-06 17:22
    I vote we ask a Mod to re-title this thread as "Chip says the P3 is gonna be HUGE!!!" just so we can see the look on Ken's face when he sees it. :smiley:

    I'm really pleased that Parallax is already looking forward to the next thing. Once the P2 hits the formal release phase, and the crush for docs/tools lets up a bit, it seems natural that Chip would look at doing a process shrink while bumping-up the internals. Will it be a P2-XP (eXtreme Performance), or a P3? Or something completely different? Maybe Chip can tease us a bit more with his thoughts.

    (I do wonder if @"Ken Gracey" is reading this thread and giving serious consideration to dumping Egg Nog on his corn flakes and calling-in sick for a week. hehehe).
  • Back on-topic, the little LNA I ordered just arrived, so it looks like we'll be playing with clock speeds and spectrum analyzer tracking generators this weekend!
  • jmgjmg Posts: 15,173
    JRoark wrote: »
    Back on-topic, the little LNA I ordered just arrived, so it looks like we'll be playing with clock speeds and spectrum analyzer tracking generators this weekend!

    I think the XI feeds the internal amplifier and XO is just xtal drive, but you should be able to get a rough gain plot XI to XO too, which is useful for Xtal usage.
    It would also be interesting to know what signal levels (AC Coupled) P2 needs for reliable clocking, by MHz.
    That may be a bathtub curve: very low MHz will need some slew-rate boost, and very high MHz need to compensate for simple loss of gain.
    The PLL switch over tests, suggest P2 can hang if it gets very narrow clock signals.
  • jmg wrote: »
    JRoark wrote: »
    Back on-topic, the little LNA I ordered just arrived, so it looks like we'll be playing with clock speeds and spectrum analyzer tracking generators this weekend!

    I think the XI feeds the internal amplifier and XO is just xtal drive, but you should be able to get a rough gain plot XI to XO too, which is useful for Xtal usage.
    It would also be interesting to know what signal levels (AC Coupled) P2 needs for reliable clocking, by MHz.
    That may be a bathtub curve: very low MHz will need some slew-rate boost, and very high MHz need to compensate for simple loss of gain.
    The PLL switch over tests, suggest P2 can hang if it gets very narrow clock signals.

    That's an interesting idea. We could clock the P2 with the tracking generator output and then hook the XOut to the SA's input and compare gain and noise. This would be roughly similar to doing an "upside-down" SWR plot, but measuring gain instead of loss. We'd just run the sweep across a very wide range (1 mhz to maybe 500 mhz?) and sweep the TG very, very slowly. We could simultaneously validate P2 operation by setting-up some code to toss some bits at a pin at perhaps 1/1000th the clock rate and monitor this output with a scope.

    Does anyone know if selecting mode %01_10 via HUBSET enables the output amp on XOut? Any ideas what the roll-off on this amplifier/feedback circuit coming out off XOut looks like?
  • jmgjmg Posts: 15,173
    JRoark wrote: »
    Does anyone know if selecting mode %01_10 via HUBSET enables the output amp on XOut? Any ideas what the roll-off on this amplifier/feedback circuit coming out off XOut looks like?

    The Xtal amp is enabled by the CC bits, and is on for not 00.
    I did a simple RF probe gain test with 26MHz and 38.4MHz clipped sine sources.
    Both worked, and 26MHz was gain > 1 and 38.4MHz showed gain somewhat less than 1 (added Caps = 0pF)
    Clipped Sine Osc have a source impedance of ~ 180 ohms

  • cgracey wrote: »
    In 14nm technology you can get 2MB (2M bytes, not bits) of hub RAM in 1 square mm. 16MB would be no problem.
    My thumbnail could hold 0.2GB!! This is insanity!!
  • It turns out the cheapie LNA from Amazon is more of a negative-gain self-oscillator than an amplifier. Lol. So its going back on Monday.

    In the meantime I’m trying to fabricate something that will handle 350-500 mhz and give me some gain because the TG output is just short of what the P2 needs to reliably operate. Judging from the results I got using a capacitively coupled freq generator as the clock input, the P2 seems happiest with a swing bigger than about 1.5 volts p/p at 20 mhz while the TG output is about 1.25 v.

    I’ll keep banging on it until the beer runs out or I break it. The smart money would probably bet on the latter. :)
  • evanhevanh Posts: 15,916
    edited 2019-12-07 17:13
    JRoark wrote: »
    ... the P2 seems happiest with a swing bigger than about 1.5 volts p/p at 20 mhz while the TG output is about 1.25 v.
    I hope you have a VIO/2 centre-bias divider on the XI input. Without the XO connected up it won't just float to 50%.

  • JRoarkJRoark Posts: 1,215
    edited 2019-12-07 17:22
    evanh wrote: »
    JRoark wrote: »
    ... the P2 seems happiest with a swing bigger than about 1.5 volts p/p at 20 mhz while the TG output is about 1.25 v.
    I hope you have a VIO/2 centre-bias divider on the XI input. Without the XO connected up it won't just float to 50%.

    Yup. 1 meg to either rail was my last attempt but this value needs some refining.

    Edit: @evanh It sounds like you have some working knowledge of this beastie. Any hints on what might make a good starting point for the divider and coupling cap values? I’m basically doing a cut-and-try based on my gut. Any input appreciated!
  • evanhevanh Posts: 15,916
    Cool, come to think about it, Chip probably has that integral to the XI input anyway. That's probably why the docs say there is 1 MOhm impedance.

  • evanhevanh Posts: 15,916
    JRoark wrote: »
    It sounds like you have some working knowledge of this beastie. Any hints on what might make a good starting point for the divider and coupling cap values? I’m basically doing a cut-and-try based on my gut. Any input appreciated!
    Ha, not that good. You're well ahead of me. I just being getting all obsessive with too much spare time.

    What triggered my thinking on this one was the closeness of the voltage needed to VIO/2.

  • evanhevanh Posts: 15,916
    My experimenting is limited to overclocked crash testing without re-engineering the eval board. Using a hair-dryer and chiller ice packs from the freezer.
  • cgraceycgracey Posts: 14,155
    edited 2019-12-07 17:50
    The XI pin always reflects the opposite of the XO pin via a driven 1 megohm resistor. So, the pin should auto balance. You shouldn't need to do anything more than connect to XI via a small capacitor.
  • evanhevanh Posts: 15,916
    edited 2019-12-07 17:36
    What would happen to XI behaviour if XO was badly loaded externally?

  • cgraceycgracey Posts: 14,155
    evanh wrote: »
    What would happen to XI behaviour if XO was badly loaded externally?

    I stated things backwards. I just fixed my prior post. XI is driven negatively with a 1 megohm resistor.
Sign In or Register to comment.