Re: [w_scan PATCH] Add Brazil support on w_scan

[Mauro Carvalho Chehab <mchehab@redhat.com>]

Em 29-03-2011 17:11, handygewinnspiel@gmx.de escreveu:
>> Em 28-03-2011 14:20, handygewinnspiel@gmx.de escreveu:
>>> Hi Mauro,
>>>
>>>> This patch adds support for both ISDB-T and DVB-C @6MHz used in
>>>> Brazil, and adds a new bit rate of 5.2170 MSymbol/s, found on QAM256
>>>> transmissions at some Brazilian cable operators.
>>>
>>> Good. :)
>>>
>>>> While here, fix compilation with kernels 2.6.39 and later, where the
>>>> old V4L1 API were removed (so, linux/videodev.h doesn't exist anymore).
>>>> This is needed to compile it on Fedora 15 beta.
>>>
>>> videodev.h should have never been in there. Was already reported and
>> will be removed instead.
>>>
>>>> @@ -1985,6 +1986,10 @@
>>>>  		dvbc_symbolrate_min=dvbc_symbolrate_max=0;
>>>>  		break;
>>>>  	case FE_QAM:
>>>> +		// 6MHz DVB-C uses lower symbol rates
>>>> +		if (freq_step(channel, this_channellist) == 6000000) {
>>>> +			dvbc_symbolrate_min=dvbc_symbolrate_max=17;
>>>> +		}
>>>>  		break;
>>>>  	case FE_QPSK:
>>>>  		// channel means here: transponder,
>>>
>>> This one causes me headache, because this one has side-effects to all
>> other DVB-C cases using 6MHz bandwidth.
>>> Are there *any cases* around, where some country may use DVB-C with
>> symbolrates other than 5.217Mbit/s?
>>
>> If you take a look at EN 300 429[1], The DVB-C roll-off factor (alpha) is
>> defined as 0.15.
>>
>> 	[1] EN 300 429 V1.2.1 (1998-04), chapter 9, page 16, and table B.1
>>
>> So, the amount of the needed bandwidth (e. g. the Nyquist cut-off
>> frequency) is given by:
>>   Bw = Symbol_rate * (1 + 0.15)
>>
>> E. g. the maximum symbol rate is given by:
>>
>> 	Symbol_rate(6MHz) = 6000000/1.15 = 5217391.30434782608695652173
>> 	Symbol_rate(7MHz) = 7000000/1.15 = 6086956.52173913043478260869
>> 	Symbol_rate(8MHz) = 8000000/1.15 = 6956521.73913043478260869565
>>
>> As you see, for Countries using 6MHz bandwidth, the maximum value is about
>> 5.127 Mbauds.
>> With the current w_scan logic of assuming 6.9 or 6.875 Mbauds by default
>> for DVB-C, 
>> w_scan will never find any channel, if the channel bandwidth fits into a
>> 6MHz (or 7MHz)
>> channel spacing.
>>
>> So, the above change won't cause regressions, although it could be
>> improved.
> 
> 
> 
> If w_scan assumes that every 6MHz cable network, also outside Brazil, will use it's maximum theoretical symbol
> rate *only* like your patch was implemented, it will probably fail as well, because in practice lower values are used.

True. I suspect, however, that there aren't many Countries using 6MHz for DVB-C. The bandwidth usage
for cable generally follows the analog broadcast standard. There are two standards used for 6MHz
analog: NTSC/M (North/Central America, Japan, Korea) and PAL/M (used only in Brazil and Laos, AFAIK - 
never had the opportunity to talk with anyone from Laos to double check it). The Countries that use 
NTSC generally follow US choice for Digital (Except for Japan: I think ISDB-C is used there).

> 
> Your patch disabled the looping through different symbol rates for any 6MHz network; therefore i was asking.
> 
> So I changed it now to scan any srate for 6MHz networks, but skip over those which are unsupported by bandwidth limitation.

Thanks for appling the fix for Brazil!

I look your implementation. It looks sane, although it will increase a lot
the scanning time. IMO, for now, except if the user uses -e or -S parameters,
I would just seek for 5127 symbol rate, to speed up scanning.

> 
> 
>>
>> IMHO, a different logic could be used instead, if the user doesn't use the
>> -S parameter, like:
>>
>> int dvbc_symbolrate[] = {
>> 	7000000,
>> 	6956500,		/* Max Symbol rate for 8 MHz channel bandwidth */
>> 	6956000,
>> 	6952000,
>> 	6950000,
>> 	6900000,
>> 	6875000,
>> 	6811000,
>> 	6790000,
>> 	6250000,
>> 	6111000,
>>
>> 	/* Weird: I would expect 6086 or 6086.5 here, as the max rate for 7MHz
>> spacing */
>> 			
>> 	5900000,		/* Require at least 7 MHz channel bandwidth */
>> 	5483000,
>> 	5217000,		/* Max Symbol rate for 6 MHz channel bandwidth */
>> 	5156000,
>> 	5000000,
>> 	4000000,
>> 	3450000,
>> };
>>
>> for (i = 0; i < ARRAY_SIZE(dvbc_symbolrate)) {
>> 	if (freq_step(channel, this_channellist) / 1.15 > dvbc_symbolrate[i])
>> 		next;
>>
>> 	/* Scan channel */
>> ...
>> }
>>
> 
> Doesnt look reasonable, since every DVB-C scan would take easily several hours this way.

True. One way to speed up would be to assume that all DVB-C channels have the same symbol rate,
by default (and provide a parameter to override it). Something like:


Probably, the easiest way to do it is:

+static int assume_all_channels_use_same_symbolrate = 1;
...
                if (__tune_to_transponder (frontend_fd, ptest,0) < 0)
                        continue;

+		if ((test.type ==  FE_QAM) && assume_all_channels_use_same_symbolrate) {
+			info("Assuming that all channels use %.04f Msymbols/s symbol rate.\n"
+			     "Use -A parameter to force testing for all symbol rates on all channels\n",
+			     sr_parm/1000000.0);
+			dvbc_symbolrate_min = dvbc_symbolrate_max = sr_parm;
+			assume_all_channels_use_same_symbolrate = 0;
+		}
...
+                switch (opt) {
+               	case 'A':
+				assume_all_channels_use_same_symbolrate = 0;
+				break;

(code untested)

Another optimization for speeding it up for DVB-C would be to have the symbolrate at the external loop.

> 
> Sorting into groups makes sense, one per bandwidth, but with the widely used first for each group is better. And still limit to common used srates -> done.
> 
> 
> So, it's included now, but i changed the whole stuff a little. 


Anyway, I'll test the today's version and reply if I detect any troubles on it.

Thanks!
Mauro