On 06/21/2017 06:50 PM, Daniel Scheller wrote:
Am Wed, 21 Jun 2017 09:30:27 +0300
schrieb Antti Palosaari <crope@xxxxxx>:
On 06/20/2017 08:45 PM, Daniel Scheller wrote:
From: Daniel Scheller <d.scheller@xxxxxxx>
Add signal-to-noise-ratio as provided by the demodulator in decibel scale.
QAM/DVB-C needs some intlog calculation to have usable dB values, OFDM/
DVB-T values from the demod look alright already and are provided as-is.
Signed-off-by: Daniel Scheller <d.scheller@xxxxxxx>
---
drivers/media/dvb-frontends/stv0367.c | 33 +++++++++++++++++++++++++++++++++
1 file changed, 33 insertions(+)
diff --git a/drivers/media/dvb-frontends/stv0367.c b/drivers/media/dvb-frontends/stv0367.c
index bb498f942ebd..0b13a407df23 100644
--- a/drivers/media/dvb-frontends/stv0367.c
+++ b/drivers/media/dvb-frontends/stv0367.c
@@ -25,6 +25,8 @@
#include <linux/slab.h>
#include <linux/i2c.h>
+#include "dvb_math.h"
+
#include "stv0367.h"
#include "stv0367_defs.h"
#include "stv0367_regs.h"
@@ -33,6 +35,9 @@
/* Max transfer size done by I2C transfer functions */
#define MAX_XFER_SIZE 64
+/* snr logarithmic calc */
+#define INTLOG10X100(x) ((u32) (((u64) intlog10(x) * 100) >> 24))
+
static int stvdebug;
module_param_named(debug, stvdebug, int, 0644);
@@ -3013,6 +3018,33 @@ static int stv0367ddb_read_status(struct dvb_frontend *fe,
return -EINVAL;
}
+static void stv0367ddb_read_snr(struct dvb_frontend *fe)
+{
+ struct stv0367_state *state = fe->demodulator_priv;
+ struct dtv_frontend_properties *p = &fe->dtv_property_cache;
+ int cab_pwr;
+ u32 regval, tmpval, snrval = 0;
+
+ switch (state->activedemod) {
+ case demod_ter:
+ snrval = stv0367ter_snr_readreg(fe);
+ break;
+ case demod_cab:
+ cab_pwr = stv0367cab_snr_power(fe);
+ regval = stv0367cab_snr_readreg(fe, 0);
+
+ tmpval = (cab_pwr * 320) / regval;
+ snrval = ((tmpval != 0) ? INTLOG10X100(tmpval) : 0) * 100;
How much there will be rounding errors due to that signal/noise
division? I would convert it to calculation of sums (tip logarithm
calculation rules).
This is taken from stv0367dd aswell, the reported and calculated values are in 0.1dB precision. This and to not diverge any more from the "source" driver, I'd prefer to keep it how it is. These are just simple tuner cards anyway and by no means professional measurement gear, and should only give a more or less rough estimate on reception quality. E.g. my stv0367 cards report around 36dB SNR, whereas the cxd2841er reports ~37dB, compared to my DOCSIS modem, which reports 34dB on DOCSIS channels (another variant I had earlier even reported 39dB on the same channels), so... Even, we get way more precision than on the relative scale calc on the cab_read_snr functions which is in 10%-steps...
Also, that INTLOG10X100 is pretty much useless. Use just what
intlog10/intlog2 offers without yet again another conversion.
Will check and experiment. Again, taken from stv0367dd :-)
You should understand that there is no floating points on kernel, thus
that kind of divisions needs special attention. It should be written
log10(signal) - log10(noise) in order to minimize rounding errors. Lets
say as example if you divide 2 by 3 you will get 0, not 0.666... So
depending on actual numbers used on calculation, there is more or less
rounding errors which are easily avoidable.
Antti
--
http://palosaari.fi/
