Silicon delay lines

Good practice is anything that works.

Delay lines can be very handy. The LS31 is pretty sloppy, but there are more precise parts around, like the Dallas thing... Bel Fuse and Data Delay have some too. Micrel makes an astonishing ECL delay line, SY89295, which has 1024 internal 10 ps delay elements and guaranteed monotic setability.

We program delay lines inside FPGAs now and then; you have to fight the Xilinx tools to do it. Some of the new Vertex chips include legally-sanctioned delay blocks.

All-synchronous design is sort of a severe ancient religion.

John

Guess so, as long as you know what you're getting into. I've got this personal thing that's been bugging me like a splinter under a fingernail for a while. I'd like to try a delay line, but I'm wary of things I don't understand. Want to learn, you see.

If only Digikey carried them. I hate specifying parts that other people will have trouble ordering. And I don't want to be in the parts business.

Ah yes, Micrel. A bit too much for a 32MHz CMOS clock...

No PLDs in my design. Just a few flip-flops and a PIC.

Heh. I'm planning on some pretty good power supply bypassing on the delay line chip. I understand there's some kind of ramp/comparator thingy in there. I'm hoping that overall it's less sensitive to power supply noise than the VCO I've got right now. Would you heavily filter the VCC to a delay line?

I never do and have designed some out. Got less noise that way and the BOM cost total plummeted impressively. The latter made that client really happy. They were kind of expecting that I'd solve their noise issues and later didn't really want to know why the noise was now gone. But they sure didn't expect a serious cost reduction, especially because it just "happened" they didn't have to pay extra for that service.

If, for example, there had to be a precise summation of signals over staggered delays I used Belfuse LC lines as John mentioned. For anything else it's the good old LC, RC, LR, whatever is practical. IOW nickel and dime parts. Then again my background is analog so maybe I am a bit biased here.

Sure... I used some DS1000-series delay line (and some logic) once to create a double-speed clock to extract bytes from a 16-bit word-wide parallel data bus. The clock speed was fixed, the DS1000 part had decent tolerances, so after sitting down and checking the timing margins, it really was a solid, repeatable design.

In fact, the main advantage of something being sold as a delay line over just using some handful of spare gates is -- generally -- the much tighter tolerance you get on the delay.

"NOTIFICATION OF DISCONTINUANCE END OF LIFE: DELAY LINES ALL PRODUCT FAMILIES

Bel is issuing this Notification of Discontinuance regarding the End of Life status of our delay line products. This unilateral discontinuance is in support of rather than in lieu of any prior notification you may have received from Bel. Original notification dates are in no way superceded by this announcement."

Oh dear. I suspect 100000 transistors on an ASIC are cheaper than a LC delay line.

Yeah, mine had been a long time ago. They were used in ultrasound beamformers (that's what I often design for a living). Then, with digital beamforming becoming economical the market for delay lines has shriveled to the point where it just doesn't make sense anymore. But you can still get similar delay lines from others. Be prepared for sticker shock though. IMHO these have become boutique parts. While it may be ok to use one in a rocket controller it might not be the right thing to do in regular electronics.

As I wrote, personally I'd stay away from a "canned solution" and roll my own. It ain't that hard. Heck, we even roll our own pizza dough instead of buying anything "ready to bake" :-)

Besides noise and a painful sticker price there is another reason why I avoid commercial delay lines if at all possible: Neither I nor my clients like single-sourced parts. Over my career I have heard the clanging of too many last order bells and seen the demise of whole companies along with their otherwise superb products. Remember Plessey and their excellent mixer chips? Luckily I was able to secure a small stash of those before the ship sank.

Monostables have ramps and comparators and can have nasty sensitivities to noise. Some of the Dallas parts that Dallas list as delay lines are actually strings of monostables.

Proper delay lines don't have ramps and comparators, and are much less noise sensitive. I've used them in a few applications, but they do tend to come out expensive.

Farnell stocked the Newport lumped constant delay lines for some twenty years - they were still in the catalogue last year, albeit as C&D Technology parts - but they've vanished from the 2007 catalogue.

These parts were just passive linear phase phase low pass filters, built as thick film hyrids, and - as passive parts - they were totally insensitive to power supply noise. If you get hold of a book on filter theory, you can design your own delay line as an "all-pass" filter, but you get twice as much delay out of the same capacitance and inductance if you design your delay line as a low-pass filter. You do have to make sure that the high frequency cut-off of you low pass filter doesn't excessively slow down the transition times of the wave- train you plan on delaying.

-- Bill Sloman, Nijmegen

You can still get those types of delay lines today:

Digikey has that series in stock. But sit down before looking to the right where the prices are. Big bucks. Key "Susumu" into search, then first item under "Inductors".

It's the other way round, you get twice the delay of the low-pass version using the components as an all-pass. You probably wouldn't like the shape of the output edge, though;-)

Regards Ian

That's not the way it seemed to be working when we looked into this at Cambridge Instruments around 1985 - which is why we went ofr the low- pass filter.

A phase linear low pass filter may degrade the steepness of the output edge, but since the delay will remain constant with frequency up to the cut-off point, it won't degrade the shape.

A phase linear all-pass filter wouldn't - in theory - have a cut-off point, so it ought to be better.

-- Bill Sloman, Nijmegen

For a normalized 1st order APF Hap= (1-p)/(1+p) For a first order LPF Hlp= 1/(1+p)

The APF has twice the LPF phase shift for the same cut off frequency.