[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: How ensure consistency with timing signals
From: |
Marcus D Leech |
Subject: |
Re: How ensure consistency with timing signals |
Date: |
Wed, 26 Feb 2020 15:25:41 -0500 |
Something to understand about the throttle block is that it doesn’t guarantee a
precise sample rate. Just that the long term average rate is what you expect.
Sent from my iPhone
> On Feb 26, 2020, at 12:09 PM, Lukas Haase <address@hidden> wrote:
>
> Hello,
>
> I use a square wave to generate control/clocking signal for some control
> logic:
>
> https://snipboard.io/WQ8p6d.jpg
>
> The work function of the "Controller" block looks like:
>
> def work(self, input_items, output_items):
> d = np.diff(input_items[0])
> inds = np.where(d == 1)
> for ind in inds[0]:
> nedge = self.nitems_read(0) + ind
> delta = nedge - self.nedge_prev
> print("EDGE @ " + str(nedge) + ", Delta=" + str(delta))
> self.nedge_prev = nedge
> return len(output_items[0])
>
> I.e., it detects the rising edge. The code above stores the previous edge and
> shows the delta between them.
> If the frequency of the square wave exactly divides the sampling rate, I
> would expect that this delta is always constant/consistent (and I need this
> for my application).
>
> However, with samp_rate=5M and freq=10:
>
> EDGE @ 10751169, Delta=500054
> EDGE @ 11251224, Delta=500055
> EDGE @ 11751278, Delta=500054
> EDGE @ 12251332, Delta=500054
> EDGE @ 12751387, Delta=500055
> EDGE @ 13251441, Delta=500054
> EDGE @ 13751496, Delta=500055
> EDGE @ 14251550, Delta=500054
>
> One interval is not only not always constant but is 54/55 samples larger than
> the expected 500000.
> When making the interval small (e.g., samp_rate=5M, freq=1k, it works:
>
> EDGE @ 65072534, Delta=5000
> EDGE @ 65077534, Delta=5000
> EDGE @ 65082534, Delta=5000
> EDGE @ 65087534, Delta=5000
> EDGE @ 65092534, Delta=5000
> EDGE @ 65097534, Delta=5000
> EDGE @ 65102534, Delta=5000
>
> I tried tried offsetting the frequency by a slight but but no change. E.g.,
> freq = 1.0 / (1.0/10 - 1.0/samp_rate)
>
> I know I could implement a logic that adds a constant directly in the work
> function but to keep things manageable I would really prefer to use clock
> signals (there will be a bunch of them floating around in the end).
>
>
> Thanks,
> Lukas
>
>
>
>