Bit of IQ

That circuit, with design tables, is in the Williams filter book. Unfortunately, the outputs differ by 90 degrees but are not +-45 relative to the input. They squirm all over the place.

I guess it's like the phase sequence filter (PSF). That just keeps wrapping the phase around and around... The step response (of PSF) is a weird beast.

George h.

On a sunny day (Thu, 30 May 2019 07:50:49 -0700) it happened John Larkin wrote in :

Yes, it shifts a bit, but for audio it makes litte difference I think. Nice circles on the analog scope :-)

On a sunny day (Thu, 30 May 2019 17:25:05 GMT) it happened Jan Panteltje wrote in :

PS you can make those test sweeps in Linux with 'sox':

create sweep wave file: sox -V -r 48000 -c 2 -b 24 -n sweep.wav synth 10 sine 240:2400

continuously play sweep wave file: while [ 1 ] ; do play sweep.wav ; sleep 1 ; done

or create one tone: sgen -s 48000 sin 2400

If you look at the xy circle on the scope you will easily see any phase or amplitude changes.

To be causal, the output has to be delayed from the input at least the time of a quarter wave at the lowest frequency to be handled, usually longer.

For audio, or SSB generation, absolute phase doesn't matter. We make an IQ modulator box to simulate eddy-current transducers, and absolute phase "absolutely" matters.

The box that makes a true wideband 90 degree shift is a Hilbert transform, which sadly is impossible. It's non-causal, like an ideal lowpass filter; its output impulse response begins before the input.

You can make narrowband ones that work pretty well. The major problem with wideband ones is that both the central spike and the long tail of the impulse response have infinities.

Cheers

Phil Hobbs

Like a lowpass, you can approximate the ideal transfer function, but it adds delay. That's an oxymoron for a phase shifter.

We could cheat and add a PLL or something to shift our "RF" input, assuming that a transducer excitation is going to be pretty constant.

WTF ?

..... Phil

Or else calibrate the filter and fix it in software.

Cheers

Phil Hobbs

Huh, the impulse response... I'll try that next time I'm testing a lockin. (where I use the PSF... missing a few cycles in a lockin hardly matters if you're averaging over hundreds. :^)

George H.

One of the things I like about our I/Q box is that it's 100% analog. That's nice once in a while.

About 50% of our output is all-analogue, but that number is going down a bit on account of auto-tweaking and interface issues. We really like the LPC804 and LPC845 Cortex M0+ for low-level stuff. The 804 has a reasonable amount of FPGA fabric--about the same as a CPLD costing twice as much, not even counting the M0+ core.

Cheers

Phil Hobbs

On Thursday, May 30, 2019 at 1:46:14 PM UTC-7, Phil Hobbs wrote: [about 90 degree phase shift]

If you're using a narrowband source, an LC tank oscillator for instance, a low-Z probe (current transformer) on current and high-Z probe (FET follower) on voltage gives you the two phases. No filter required, no time delay.

The logical problem, is that a wideband source (white noise) has no well-defined phase until you sample it for a while, then FFT it. Then call the resulting mish-mash a 'spectrum' for which each frequency has an associated phase. Time delay is well-defined on a random source, but phase shift isn't.

All links working OK here (Linux Mint & Pale Moon browser). Are you getting an error message or just nothing happening?

There's not much logic in that one!

We're about ready to pick a new generation of uPs and FPGAs, and there are lots of opinions here. I'm leaning towards using ZINQs. We already use the 7020, a $125 monster with two ARM cores that usually runs Linux. We need a lower-end part, and I'm thinking Zynq 7007, 400 balls, single 760 MHz ARM core and 23K logic cells. That's under $50 in quantity. Hardware float is nice.

Some people here favor a soft core CPU inside an FPGA, but that needs a new tool chain and gobbles valuable FPGA RAM. No hardware FP, probably.

Lattice and Altera have some low-end SOCs too. We tried to run the Lattice tools but couldn't get past FlexLM.

I'm using an RC 45-deg lead into one side of a diff-in multiplier, and another RC 45 lag into the other. Four parts makes 90 degrees shift and gain=1. It's pretty good over a modest frequency range.

fredag den 31. maj 2019 kl. 19.55.48 UTC+2 skrev John Larkin:

those prices sound high, you can get a whole board for about that

they don't have any in stock but LCSC quoted us Zynq 7020 for about half of the digikey price

No, it's just a couple of dozen LUTs, but that's enough to offload some task that needs hardware timing and is annoying to do with counter/timers--stuff where several things need to happen with very consistent timing, and you'd otherwise need external glue logic.

ARM Cortexes have good interrupt timing, but their latency varies a bit--12 cycles from mainline code, but only 8 from another handler.

The 804 doesn't have that much flash, so you can't use it for super complicated stuff, but it's a very nice chip for what it is.

IIRC there's a full open-source toolchain for Lattice.

Cheers

Phil Hobbs