<azonenberg>
The rolloff is a bit slow so you might get a bit of aliasing in 4 channel mode with wideband signals (125 MHz is Nyquist and it's only down about -5.5 dB there)
<azonenberg>
but it's down -24 dB at 250 MHz
<electronic_eel>
lgtm, but I don't do filter design often, so I tend to have to read up on it each time
<azonenberg>
again it's a tradeoff. I don't want to sacrifice 1-channel performance to remove all chance of aliasing on 4-channel mode
<electronic_eel>
yes, the tradeoff looks sensible to me
<electronic_eel>
and no peaking issues
<azonenberg>
also super linear phase
<azonenberg>
thats the thing about bessel filters, they have slower rolloff but trade that for much smoother phase
<electronic_eel>
phase problems are hard to fix in software
<azonenberg>
Yeah
<awygle>
i always get a warm glow when people take my advice :p
<azonenberg>
awygle: i think that glow is called "cerenkov radiation"
<awygle>
i usually recommend odd-order filters
<azonenberg>
awygle: oh?
<awygle>
it's not a huge deal
<awygle>
it just makes them symmetrical
<azonenberg>
What's the benefit of that?
<azonenberg>
signal processing of that sort is not my strong point
<azonenberg>
looking at the group delay plot, the 7th order bessel filter has significantly less nice looking group delay/phase
<azonenberg>
the 5th is ok but has slower rolloff
<awygle>
there's a DFM element to it having to do with component values
<awygle>
and also if you care about whether it's a short or an open at DC
<azonenberg>
This is a LPF, and yes it needs to work down to DC
<awygle>
if you pick an even-order filter you get a short at one end and an open at the other
<awygle>
for some topologies
<awygle>
shouldn't be an issue here
<awygle>
just a rule-of-thumb marker fired in my brain
<azonenberg>
yeah this filter design has 11 nH series and 13 pF shunt
<azonenberg>
then 51 nH / 27 pF
<azonenberg>
and 91 nH / 75 pF
<Degi>
I didn't really read the whole convo and not sure if that was said, but if you use a relais with two poles, you could use 1 relay to switch a filter in and out (unless you do differential)
<Degi>
Also is it possible to use DSP to apply the inverse of the filter function to the signal?
<azonenberg>
Degi: you don't want to invert the antialiasing filter, but inverting the transfer function of the gain/offset stages is very much possible and something i'm looking into adding in glscopeclient down the road
<azonenberg>
i will be doing probe de-embedding at some point too
<Degi>
Hm you could fix the phase response of the antialiasing filter
<azonenberg>
The phase response of the AA filter is super linear so i dont think it will be necessary?
<Degi>
Okay
<Degi>
And amplitude ripple is good too?
<azonenberg>
group delay is 3.895 to 3.781 ns across the DC - 100 MHz band
<Degi>
Hm like if the phase near max frequency is different from phase at lower frequencies, that would still lead to square waves being distorted
<azonenberg>
that's the advantage of a bessel-thomson filter. Slower rolloff than other filters but very smooth
<azonenberg>
no ripple in amplitude, just a slow gentle rolloff to -3 dB at 100 MHz
<Degi>
Huh good group delay
<azonenberg>
I'll have an updated schematic for you all to look at shortly. Designing the final gain stage now
<azonenberg>
at that point all i'll have left to do is the PSU and MCU
<azonenberg>
If you've got some time, start thinking about how we're going to power this. I don't need a full schematic design done, just ideas on what chips to use
<Degi>
For the buck and negative rails etc?
<azonenberg>
I need to go from +12V input to +/- 5V and +/- 6V
<Degi>
I'm not very good at this but I know who to ask...
<azonenberg>
as smooth as possible
<azonenberg>
so i'm thinking probably smps to +/- 6.5-7V or so then LDO to the final voltages
<Degi>
What do we have as input? 12V?
<azonenberg>
Yeah
<azonenberg>
let me get you some load estimate
<azonenberg>
+/- 6V is only the offset stage, call it 30 mA on each leg
<azonenberg>
-5V is used by the final output amplifier, call that 30 mA as well
<azonenberg>
+5V is used by the output amplifier (30 mA) and the gain stage, which is much more hungry - budget 200 mA for that
<Degi>
Hm I've MSGed that person, but I'll search thru mousre some
<Degi>
And we need something for the HMCAD too, 1.9 V or so
<azonenberg>
no
<azonenberg>
this is an AFE charaterization board only, there's no ADC on it
<Degi>
Ah right
<azonenberg>
i intend to reuse as much of the design as possible on the board with the ADC of course
<azonenberg>
oh, then the LM393 needs a mA or so
<Degi>
She says to use inverting buck boost for negative rail
<azonenberg>
Yes, that much is certain
<azonenberg>
the point is more exactly which chip :p
<Degi>
Yes
<azonenberg>
And we need 3.3V for the MCU as well. Budget 30 mA for that
<Degi>
Hm does that need to be good
<electronic_eel>
azonenberg: if you want a really low noise post-switcher ldo combo for +- rails: ti tps7a49 and tps7a30
<azonenberg>
electronic_eel: will look into it
<azonenberg>
oh, my other not-hard but nice requirement
<azonenberg>
i prefer qfn/dfn/bga packages to soic/tssop/tqfp
<Degi>
Ok, why is that?
<azonenberg>
I've had better assembly yield, they're also smaller so you can fit more channels in the same board space
<Degi>
Yup LDOs available in DFN
<Degi>
What should I do, components? Schematic? Board layout?
<azonenberg>
many of my boards are 100% leadless, i only use packages with pins if i can't find the part i need in a DFN/QFN package
<azonenberg>
Just find some parts that look decent for the moment
<azonenberg>
multiple people on one kicad design tends to be difficult
<Degi>
Oh no, TI seems to have problems
<azonenberg>
?
<Degi>
Nvm my browser has problems
<Degi>
How precise should the rails be?
<azonenberg>
+/- 6.6V is absolute max for the LMH6552, +5.4 for the 5205. I would target +/- 100 mV of nominal voltage, ideally a bit tighter than that
<azonenberg>
but what matters more for low noise is ripple
<azonenberg>
if my 6.0V rail is steady 5.95 or 6.05 that's not a problem, if it's fluctuating between them it is
<Degi>
Okay
<azonenberg>
I'll do the detailed psu design later, just looking for suggestions on parts to use at the moment. Being in stock at digikey is a strong plus
<electronic_eel>
for positive supplies there are some lower noise ldos available than the tps7a49, for example the LT3042
<electronic_eel>
but for negative supplies the field gets small
<azonenberg>
Yeah i figured
<electronic_eel>
and it won't help much if you have a super positive ldo, but the negative one sucks
<electronic_eel>
you want to trigger on a bit more than 5v, correct? how about 13k and 1.5k -> 0.517v
<Degi>
Yup LT3093 seems to be one of the best for negative
<electronic_eel>
now when looking at the overvoltage shutdown I'm missing a latch function:
<electronic_eel>
when the voltage just goes above 5.1x v the relay will switch off, if the voltage drops, the relay will switch on again
<azonenberg>
electronic_eel: yes, my thought was to have the MCU be monitoring that signal
<electronic_eel>
this can lead to lot's of relay chatter and ruin the relais
<azonenberg>
if it goes low, start driving low
<azonenberg>
i.e. software latch
<electronic_eel>
I'd prefer a hard logic latch
<azonenberg>
then after say 5 sec release its driver and see if the input shuts down again or not
<electronic_eel>
that switches off the relay without software intervention
<electronic_eel>
and a reset function from the mcu to reset the latch
<azonenberg>
well i'd probably do this in FPGA in the final system
<Degi>
Hm yes maybe a hardware RS latch resettable and settable by software and the logic
<azonenberg>
This is a characterization board
<electronic_eel>
you can use a 74lvc1g14 for the latch
<electronic_eel>
very small and cheap
<azonenberg>
kk
<electronic_eel>
here is a very fancy version of that from a project I'm working on right now: https://imgur.com/a/1DyGyHe
<electronic_eel>
it has a configurable rc delay stage in front
<electronic_eel>
with the comparator and voltage divider, you are better off to do a delay with caps after the voltage divider before the comparator
<electronic_eel>
so you can skip the part left of the 10k series resistor in my schematics
<electronic_eel>
oh, I just see that I'm better off using one 74lvc2g14 instead of two 74lvc1g14
<electronic_eel>
when you are sure your mcu doesn't do any stupid stuff during reset or bootloader or whatever, you can skip the transistor, directly connect the mcu to the '2g14 and configure the pin as open-drain
<awygle>
the physical power switch inside my bluetooth headphones is cracked