Date: 10 Jun 2015 14:38:54 -0400
Message-ID: <20150610183854.19549.qmail@ns.horizon.com>
From: "George Spelvin" <linux@horizon.com>
To: arjanvandeven@gmail.com, linux@horizon.com
Subject: Re: Discussion: quick_pit_calibrate is slow
Cc: a.p.zijlstra@chello.nl, adrian.hunter@intel.com, ak@linux.intel.com,
        akpm@linux-foundation.org, arjan@infradead.org, bp@alien8.de,
        hpa@zytor.com, linux-kernel@vger.kernel.org, luto@amacapital.net,
        mingo@kernel.org, penberg@iki.fi, tglx@linutronix.de,
        torvalds@linux-foundation.org
In-Reply-To: <20150610162738.24160.qmail@ns.horizon.com>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2195
Lines: 55

George Spelvin wrote:
> spread spectrum clock modulation rates are typically about 30 kHz

Actually I found a source:
http://download.intel.com/design/Pentium4/guides/24920601.pdf
CK00 Clock Synthesizer/Driver Design Guidelines

Page 45 says
"8. The modulation frequency of SSC is required to be in the range of
    30-33 KHz to avoid audio band demodulation and to minimize system
    timing skew."

> The faster the modulation rate, the more the EMI peak is flattened.
> But too fast produces unaceptable cycle-to-cycle timing variations,
> also known as clock jitter.

This is actually wrong.  It's the modulation *depth* that determines the
peak reduction.  (The above Intel spec says 0.5% typ, 0.6% max.)
The only requirement is that the rate is well below the clock rate,
and well above the averaging time of your frequency analyzer used to
measure the noise.

But basically, they want the modulation rate above the audio band, but
as low as possible to make it easy for downstream PLLs to track.


To quantify the error, consider a triangular modulation spectrum with a
total of 0.6% of range.  Intel requires <= 0.6% = 6000 ppm of down-modulation
relative to nominal, but I'll consider it to be +/-3000 ppm from an
average.

The total phase error accumulated during a frequency excursion is the
integral, which is the area under the peak.

(The Lexmark/Hershey's kiss modulation waveform, which provides a slightly
flatter peak, has a slightly smaller area due to the curved edges, so
triangular is a safe overestimate.)

At 30 kHz, during one peak (1/60 ms = 16.66 us), the oscillator phase
advances by 16.66 us * 1500 ppm = 25 ns.  (At the more typical 0.5%
value, that's 21 ns.)

25 ns is 100 ppm of 0.25 ms, so it should be okay if I go use a measurement
interval of 1 ms or more.


Some BIOSes have offered 1% spread:
http://www.techarp.com/showFreeBOG.aspx?bogno=266

which doubles that figure, but I don't think there's anything more.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/