spud semi

Am 23.07.22 um 19:03 schrieb snipped-for-privacy@highlandsniptechnology.com:

Oh, the first reference to potato semi that is not from a high end audio web site, and with ebay as distributor.

Sorry, no need for inverters that cannot drive 30% of their own input capacitance with the delays given. (last time I looked, a few years ago)

Gerhard

Fun.

The datasheets all seem to be from 2010, and Octopart doesn't return anything for a search on 'PO74'. A pity. I could certainly use a noiseless 1.25 GHz HC4046, for instance. (Many extra points for fixing the oscillator nonlinearity, some extra points for moving the deadband away from the servo point.)

Cheers

Phil Hobbs

It seems, the 4096 with it's cleaned-up deadband has died.

Gerhard

Yeah, we've gone round the mulberry bush about the 4046's deadband several times here in the last decade or so. It's no big deal as long as you know the trick--a 1M resistor to ground, just enough to pull the servo point a few nanoseconds to one side.

A 1.25 GHz version would be pretty slick, in a 1980s retro sort of way. ;0

Cheers

Phil Hobbs

Which phase detector do you prefer ?

boB

I haven't built an RF PLL in a few years, but at VHF and above I normally use a diode ring mixer such as a Mini Circuits MPD-1. Next time I want a narrowish-bandwidth loop with very accurate average phase, I might try out JL's PECL d-flop trick.

Cheers

Phil Hobbs

What I needed wasn't frequency lock as such, but picosecond time alignment of my clock the the OC3 optical data stream.

Imagine using an XOR or diode mixer to compare my vcxo to an incoming square wave. Period is 7 ns, so a +-0.5 volt xor (or mixer) will have an error slope of 0.3 volts per ns, 300 uv/ps. Tiny analog errors anywhere make picoseconds of time error short term and long term. The ecl d-flop has an essentially infinite detector gain, and differential ecl is super temperature stable.

I analyzed the loop assuming that the effective gain was determined by jitter and guessed that the detector output would go rail to rail differential in 20 ps, once it was lowpass filtered. That was the starting point for tweaking, but it was about right. There are a few papers out there on infinite-gain pll's but they mostly confused me.

Simple phase detectors don't find lock if the VCO isn't already close to the right frequency, so I used a switchable loop filter, wideband to find lock then narrow for low jitter.

Yup, that's pretty compelling. AIUI the loop bandwidth has to be fairly narrow, though, because there's a large-amplitude pseudorandom pulse train coming out of the d-flop that you have to get rid of.

Yeah, and pull-in is unreliable if the loop filter isn't super simple.

When the loop is out of lock, the phase detector produces a beat note at the difference frequency. The VCO frequency gets pushed slightly higher on the positive half cycle of the ripple, and slightly lower on the negative half cycle. The half cycle tends to reduce the frequency error will get stretched out slightly, and hence the DC component of the beat note will not be zero, but will gradually push the loop toward lock.

With a one-pole loop filter, i.e. a lead-lag integrator, the small average DC will push the loop towards lock until it gets within the closed-loop bandwidth, at which point it jumps into lock. This mechanism is called "pull-in". It tends to be quadratically slow--if your initial error is 10x the closed-loop bandwidth, it takes four times longer to lock than if it's 5x. (Every acquisition transient is an individual, of course, but the tendency is quadratic.)

If you put in additional poles to help get rid of the reference frequency ripple, they phase shift the reference ripple, so that there are beat-frequency regions where the pull-in voltage changes sign, tending to push the loop further from lock. Once that starts happening, the loop drifts towards the nearest stable null (where the pull-in voltage passes through zero). This condition is called "false lock".

Thus it's usually helpful to add some acquisition aid. One can use a phase-frequency detector, for instance, or (my fave) a triangle sweep made out of the loop integrator plus a Schmitt trigger, e.g.

Cheers

Phil Hobbs

That system sends Manchester data at 77 Mbits/sec. It sends a square wave most of the time, with burst of data at about 22 KHz, a couple hundred bits. This PLL actually uses the data edges to make lock decisions too.

There are ambiguities so we may have to flip the 77 MHz clock phase and try again.

The wobble-towards-lock is fun to see, but as the VCO freq is further off, the wobble gets weaker and eventually doesn't work.

With 2 KHz loop bandwidth and a 77 MHz clock, ripple wasn't a problem. VCXOs have their own lowpass filters inside, ahead of the varicap, typically in the 10s of KHz, often undocumented.

All great fun. I had software looking for broken-lock indicators and switching the clock phase and the loop filters, which was good enough, given a perfect GPS based data stream and a good, expensive VCXO that was pretty close to start.

We've had a couple hundred in the field for about 20 years and they have been very good. Too good. I'm trying to convince the user that it's time to replace them all, but the dumb things refuse to wear out.

Like old HP laser printers. ;)

Cheers

Phil Hobbs

Mine did, and the cartrige prices were absurd anyhow. I got a Brother that's great. Refills are big and cheap, and it prints on both sides, which I didn't expect from a cheap printer.

It just sits there quietly. The HP used to power itself up and do a cleaning or something noisy in the middle of the night.

We have a LaserJet 2300DTN, which does duplex as well. It's recently become a bit hard to get good cartridges for it, so it may have to be replaced. :(

Cheers

Phil Hobbs

My venerable HP LJ6L only died a couple of years back, after well over a decade of service, including many toner transfer PCBs. When I dismantled it I found no fewer than eight things broken internally, yet none had stopped it from working until the last one.

The replacement HP P2055dn is nice with the duplexer and all, but it has such advanced toner management (turning solid blacks into microscopic cross-hatch patterns) that it's useless for toner transfer, so it's just as well the PCB prototyping services are so cheap and quick these days. I still miss being able to do a Saturday afternoon turn-around from layout to soldering an under an hour though.

Clifford Heath

HI Clifford, Re cross hatch pattern, maybe it is in "toner-saving" mode, a.k.a. draft mode. If even "standard" quality has this problem then try "high-quality" mode. (I couldn't confirm that this model has those quality settings, as strangely, HP does not have any specs for it posted on their Support area.)

In general monochrome lasers from HP & Brother have had the fewest reported major problems ('more reliable') than any other brand or type of printer. If queried about what printer to buy, always start your recommendations there, friends. Regards, Rich S.

I think I've tried all the settings hat are available to me through MacOS CUPS printer subsystem, and through the printer menus. It's possible I missed something of course. But "solid" blacks were always difficult for laser mechanisms, because to evenly distribute a strong static charge (without it repelling itself and causing white blotches) is hard. So I think the cross-hatching is a feature of how they handle that.

I also acquired a HP Color C1518ni (best possible price - I swapped it for a homegrown pumpkin!), but only turn it on when I want colour, because the cartridges are so expensive.

A lot of folk who've tried toner transfer with Brother lasers say that the toner doesn't transfer. I don't know what series or version those were, but I didn't want to buy one to find out that it didn't work.

Clifford Heath.