<pie_>
<qu1j0t3> that will be more true once we have proper public access to the last 1000 years of published research, but sure
<pie_>
to be fair im not sure what one could say about smaller universities :P
<qu1j0t3>
they still have subscriptions and such though?
<qu1j0t3>
and SOME library
genii has quit [Quit: GO LEAFS GO !]
Bike has joined ##openfpga
<awygle>
i feel like smaller universities usually have a decent network for inter-library loan too
<qu1j0t3>
decentralisation is such a good thing
<qu1j0t3>
several countries have Federal university systems and it's such a good idea
<kc8apf>
slightly more 7-series docs up on prjxray.rtfd.org
<cr1901_modern>
qu1j0t3: "Pineapple on pizza is okay actually" ... is that a fact?
<qu1j0t3>
try it and see!
Dolu has quit [Ping timeout: 265 seconds]
pie_ has quit [Read error: Connection reset by peer]
pie_ has joined ##openfpga
pie_ has quit [Excess Flood]
pie_ has joined ##openfpga
m_w has quit [Quit: Leaving]
noobineer has joined ##openfpga
pie_ has quit [Ping timeout: 256 seconds]
rohitksingh_work has joined ##openfpga
digshadow has quit [Ping timeout: 240 seconds]
digshadow has joined ##openfpga
azonenberg_work has quit [Ping timeout: 252 seconds]
noobineer has quit [Ping timeout: 265 seconds]
<whitequark>
rqou: you need to sleep less AND to take more stimulants
<rqou>
wat
<rqou>
what's the context for that comment?
Lord_Nightmare has quit [Ping timeout: 248 seconds]
Lord_Nightmare has joined ##openfpga
<rohitksingh_work>
whitequark: Hi! I tend to sleep *so* much (more than 8 hours). Any ideas to reduce that? :p
sunxi_fan1 has joined ##openfpga
<whitequark>
rqou: 01:20 < rqou> oh btw on the topic of ENOTIME: whitequark you joked to sleep less, but i find that that makes things even worse
<whitequark>
rohitksingh_work: more stimulants?
azonenberg_work has joined ##openfpga
<sorear>
I need to try that then
<rqou>
whitequark: i'm trying to switch to biphasic sleep as awygle recommended
<rqou>
might work since apparently i've actually done that before without realizing
<rohitksingh_work>
whitequark: I'm all ears! I haven't taken any yet. Which ones would you recommend?
<rqou>
caffeine? :P
<rqou>
(not very effective for me)
<rohitksingh_work>
rqou: same for me....not very effective on me..sigh
<awygle>
something something you're just not using enough
* awygle
takes a dramatic sip of coffee
<rqou>
it can get me awake, but i won't be any more energized or productive
<awygle>
if i'm burned out in the afternoon sometimes i'll take a coffee nap. cup of coffee, 30 minutes rest/sleep, back to work. also effective on road trips.
<whitequark>
caffeine is a really useless stimulant mostly
<whitequark>
rohitksingh_work: start with modafinil
<rqou>
"Modafinil is currently classified as a Schedule IV controlled substance under United States federal law"
<sorear>
i suspect you'll find it's significantly easier to acquire in berkeley than e.g. methylphenidate
<rqou>
lol
<sorear>
since it acquired a reputation at some point as a ~ hip nootropic ~
<rqou>
i'm not very well connected with "that part" of the berkeley culture
sunxi_fan1 has left ##openfpga [##openfpga]
<rqou>
i'm apparently so innocent i haven't "even" tried "edibles"
<whitequark>
rqou: Sch IV, also known as "basically permitted"
<rqou>
so is it basically "ask your doctor about <xxx> today"
<whitequark>
rqou: what do you mean?
<awygle>
why hasn't anyone given me this drug? i've seen like four sleep doctors for most of these things it treats
<whitequark>
modafinil has no accepted medical use in the united states
<whitequark>
because united states sucks ass
<rqou>
ah, typical tagline in (us-only?) direct drug marketing to consumers
<whitequark>
rqou: i know
<whitequark>
(us-only, yes)
<rqou>
gotta have those dick pills yo :P
<rohitksingh_work>
:D
<rqou>
huh, apparently most of the direct drug marketing _isn't_ cialis anymore
<rqou>
pharma companies need to market things that are even more ~profitable~
<rqou>
wtf is wrong with US healthcare?
<awygle>
i feel like i see Lunesta ads all the time
<awygle>
which according to this wikipedia article doesn't even work but will still addict you
<rqou>
brilliant
<awygle>
whitequark: since we're talking about drugs, can you recommend a sleep aid that won't result in me sleepwalking on my broken foot?
<awygle>
(apparently ambien is strongly contraindicated for broken bones)
<whitequark>
lol ambien
<whitequark>
awygle: melatonin 0.3mg (*not* 3mg as it's often sold)
<awygle>
yeah, i have that after the whole rqou vodka thing
<rqou>
lol
<whitequark>
the what
<rqou>
that was only a hack that i used _once_
<rqou>
the janky dilution
<rqou>
before my order of "real" 0.3mg melatonin came in
<awygle>
i am under the impression that that's more for "i'm not getting sleepy when i should" rather than "i am having a hard time sleeping in this uncomfortable cast", which is more what i'm looking for
<whitequark>
awygle: oh!
<whitequark>
well...
<rqou>
awygle: do you have a better way to split 3mg into 0.3mg? i don't even own a scale that works down to that low
<whitequark>
most of the sleep aid drugs interact with one of the two sites on GABA receptors
<whitequark>
the pill is not 3mg
<whitequark>
(of course)
<rqou>
yeah, but still
<whitequark>
that said, volumetric titration is 100% fine
<awygle>
rqou: i probably would not have done it that way because 'sleep aid' + 'alcohol' sounds like trouble (even though my understanding is that's not how melatonin works)
<awygle>
also i sleep like shit if i drink
<rqou>
it's only 5ml of alcohol after dividing
<rqou>
which isn't nearly enough to get you drunk
<awygle>
"well, the bottle's already open..."
<rqou>
lol
<whitequark>
awygle: 'sleep aid' plus 'alcohol' is only bad if the sleep aid acts on GABA receptors (again)
<awygle>
actually that's not a big risk with vodka, i hate vodka
<awygle>
whitequark: right, i know it doesn't work that way
<awygle>
me at 8pm: "i'll just submit a quick patch to this library". me at 11pm: "gcc is almost finished compiling stage 2"
<azonenberg>
lol @ scrollback
<azonenberg>
meanwhile i'm playing with some DSP stuff
<azonenberg>
and realizing just how much calculus i've forgotten since sophomore year
<rqou>
lol
<rqou>
i always sucked at calculus
<rqou>
but my intuition for signals/linear systems is ok
<rqou>
what are you working on?
<rqou>
lol i'm actually failing at calculus _right now_ :P
<azonenberg>
i'm trying to create a filter that is the inverse of a first order high-pass filter with -3 dB point at 61.5 MHz
<awygle>
i love calculus. i'm crap at stats tho
<azonenberg>
As in, given a signal passed through this HPF postprocess to recover the original (assuming no noise / infinite sampling resolution for the time being)
<rqou>
wait, this system isn't minimum-phase?
<azonenberg>
minimum phase?
<rqou>
i don't think the inverse is stable
<azonenberg>
tl;dr i'm trying to build a channel equalizer
<azonenberg>
I have a signal that goes from low tens of MHz to low single digit GHz
<azonenberg>
the channel has significant rolloff on the low end
<azonenberg>
i'm approximating it as a first order LPF, the curve fits experimental data pretty well
<azonenberg>
This is postprocessing for now, not realtime, so performance isnt a huge issue
<rqou>
wait no i'm wrong
<rqou>
herp derp
<rqou>
i haven't done signals for a while either
<sorear>
the inverse of a LPF requires infinite gain at infinite frequency
<azonenberg>
sorear: this is a HPF
<azonenberg>
And i dont need performance down to DC
<sorear>
so the inverse can only be realized over a finite passband
<azonenberg>
lowest frequnecy of interest is 10 MHz
<azonenberg>
highest i guess would be nyquist which is 20 GHz
<azonenberg>
but realistically, 5-10 GHz is about the analog b/w limit
<azonenberg>
Call it 10 MHz to 10 GHz for simplicity
<rqou>
also what sorear said
<azonenberg>
My curve fitting says that the impulse response of the channel is approximately -3.8655e8 * exp(-3.8655e8 * T)
* rqou
needs to review this material
<azonenberg>
for T in seconds
* sorear
is not familiar with filters enough to know the vernacular
<azonenberg>
So now i need to create a FIR that inverts it
<azonenberg>
Can i just create a second filter with impulse response of 1/(that)?
<rqou>
i don't think a FIR filter can work
<azonenberg>
rqou: oh?
<rqou>
first of all, the naive "try to cancel poles/zeros with zeros/poles" approach has issues but i don't remember exactly what they are :P
<azonenberg>
i was hoping i could just convolve the signal with an inverse filter
<rqou>
but also a FIR filter can only have zeros at the origin
<rqou>
try an IIR filter?
<azonenberg>
meaning? (this is well past the limits of my DSP, i dont usually do more than convolving with a Gaussian lol)
<rqou>
one way to think about it is that an FIR filter can only "remember" previous values of the input
<azonenberg>
And?
<rqou>
but an IIR filter can "remember" previous values of both the input and the output
<azonenberg>
I'm OK with a filter that introduces a phase shift
<azonenberg>
if that helps at all
<azonenberg>
a significant shift i mean
<rqou>
so an IIR filter can generate an infinite impluse response because of this structure
<azonenberg>
I dont need infinite for this, do i? Can't i just truncate at some point?
<rqou>
well, maybe
<rqou>
IIR filters tend to end up smaller
<awygle>
but less stable, iirc
<rqou>
yes
<rqou>
IIR filters can become unstable but FIR filters can't
<azonenberg>
At this point in the game i dont care about speed or size
<rqou>
you need to implement them more carefully so that quantization or other errors can't stack up
<azonenberg>
I'm just trying to determine if the data can be salvaged in post or if i have to think about board-level tweaks
<rqou>
iirc byuu_san tried to do a FIR filter for audio low pass filtering and it required several hundred taps vs an IIR filtering requiring iirc three taps
<azonenberg>
i'm ok with a thousand-tap FIR if thats what it takes
<azonenberg>
again this is a PoC, i can optimize later
<rqou>
why don't you just implement it as an IIR? :P
<azonenberg>
Either way, i have to design the filter :p
<rqou>
you can always turn IIR filters into FIR filters by setting half the constants to zero
<sorear>
azonenberg: fourier transform, divide by the estimated transfer function, done?
<awygle>
"invert the frequency response" works in the ideal case, azonenberg
<rqou>
um, not at zeros
<awygle>
right
<rqou>
but yes, in the ideal case
<awygle>
you can cheat that singularity in various ways
<rqou>
hopefully not involving breaking out the bayes rule :P
<sorear>
FIR and IIR have better latency and throughput than FT methods but involve more thinking
<awygle>
to ideally approximate i think you need to know the noise level, and maybe the signal level too? but who cares
<azonenberg>
yeah i was thinking if i did any FT it would be to design the filter
<azonenberg>
not for the filter itself
<awygle>
right, FFT, invert frequency response, IFFT back
<rqou>
nah i wouldn't implement it that way
<awygle>
do some clipping at zero/infinity
<awygle>
that's the lazy way
<rqou>
find inverse frequency response, IFFT, try to truncate time-domain result and convolve
<rqou>
azonenberg: actually have you just tried this ^ naive method?
<awygle>
according to this google result calculating the LMS inverse involves "solving a Toeplitz matrix within the framework of the Levinson-Durbin algorithm" ^_^
<azonenberg>
Starting to fall asleep at the desk
<rqou>
the problem is that in general this inverse isn't possible
<azonenberg>
Soo gonna have to try tomorrow
<sorear>
cut off the input at frequences above the point where SNR < 1
<awygle>
rqou: won't the inverse of a first order LPF have a singularity only at infinity?
<azonenberg>
again, this is a HPF
<azonenberg>
not a LPF
<awygle>
o
<azonenberg>
Yes its bizarre :)
<awygle>
... your channel fades more at low frequencies?
<rqou>
is this an optical fiber?
<awygle>
are you transmitting in magnetic media or something? lol
<azonenberg>
tl;dr an amplifier that says it's for 1 MHz to 6 GHz has significant rolloff below 100 MHz
<azonenberg>
and i'm trying to avoid a board respin to fix the AFE
<azonenberg>
so i want to salvage the low frequency components
<rqou>
wait, are you applying postprocessing, preemphasis, or both?
<azonenberg>
Postprocessing
<azonenberg>
I cannot control the input signal
<sorear>
cut off the input at frequences blow the point where SNR < 1, then
<sorear>
*below
<rqou>
i don't think this is possible without more information
<awygle>
yeah i think you can just window the FT before inverting it... or something
<rqou>
if you've filtered out some information and it's below your noise level, it's gone
<awygle>
rqou: you're talking about "math possible". think "engineering possible".
<sorear>
this is about equalization, not resurrection
<azonenberg>
rqou: its not below the noise level
<rqou>
oh hmm
<azonenberg>
it's maybe 15 dB down
<rqou>
then you can probably just try the naive way i mentioned earlier
* awygle
remembers a time where he would have killed for 15 dB of margin
<sorear>
the goal is to turn an amplifier with a gradual rolloff below 100Mhz, into one with a sharp cutoff at 1-10MHz
<rqou>
design the desired inverse filter in the frequency domain and then convert to a time domain impulse response
<azonenberg>
awygle: that's not 15 dB of noise margin
<azonenberg>
i mean, my low frequencies are attenuated 15 dB with respect to the high
<sorear>
this will work fine as long as you have at least 15 dB of noise margin
<rqou>
unfortunately i'm not very experienced with actually designing systems like this
<rqou>
i only know the math :P
<awygle>
azonenberg: i understand
<rqou>
i tried to do a deconvolution once and it didn't work
<awygle>
rqou: Berkeley (TM)
<rqou>
yes
<rqou>
very
<awygle>
god why am i awake. goodnight ##openfpga
<azonenberg>
Lol i shojuld sleep too
ovf has quit [Read error: Connection reset by peer]
futarisIRCcloud has quit [Ping timeout: 256 seconds]
ovf has joined ##openfpga
futarisIRCcloud has joined ##openfpga
Dolu has joined ##openfpga
mumptai has joined ##openfpga
nrossi has joined ##openfpga
eduardo_ has joined ##openfpga
eduardo__ has quit [Ping timeout: 264 seconds]
m_t has joined ##openfpga
Dolu has quit [Ping timeout: 256 seconds]
mumptai has quit [Remote host closed the connection]
pie_ has joined ##openfpga
rohitksingh_work has quit [Read error: Connection reset by peer]
ondrej2 has quit [Read error: Connection reset by peer]
<rqou>
"The $1 million reduction in licensing and acquisitions is a necessary response to the combination of the ongoing exploitation by publishers, and the current year’s permanent reduction in campus funding."
<rqou>
i guess we really do need open access and sci-hub
digshadow has joined ##openfpga
<rqou>
also holy **** quite a few of these resources cost the university >$10k a year
mumptai has joined ##openfpga
<pie_>
meanwhile uni builds new stadium or something
<pie_>
but yeah :C
<rqou>
so i have the spreadsheet of the proposed cuts, and it really does add up to $1m
<rqou>
most of the items aren't that expensive, but adding up all the nickel-and-diming _really does_ give you $1m
gnufan has joined ##openfpga
gnufan1 has quit [Ping timeout: 255 seconds]
<awygle>
"such as faculty compensation"
<awygle>
s/faculty/administrator/
<rqou>
that's yet another different fight
<awygle>
also please explain to me how "it costs more, so we have to spend less" makes sense lol
<rqou>
"it costs more and the budget got cut so we now have to spend less"
<rqou>
although looking at the spreadsheet it seems a bunch of things that are being canceled are the dead tree versions only
<rqou>
(i don't want to post the spreadsheet publicly because they presumably set the permissions to "UCB only" for a reason)
<awygle>
yeah that's fine, i don't actually care that much lol
<rqou>
> i guess we really do need open access and sci-hub
<awygle>
we need that anyway because not everyone is at $BIGUNI