Sorry, I had brain rottage.
So, 4 dBW is a little more than 2W (unlike 40dBW, which is 10kW,
indeed). Also, small errors in the SNR formula (suddenly dropped a
-1, but that doesn't really hurt much, there).
On 05.02.2016 11:51, Marcus Müller
wrote:
Hi Daniel,
On 04.02.2016 22:49, Daniel Pocock
wrote:
To give a more specific example:
a) SDR device sampling the 2 meter band (144 - 148 MHz), this input
range is locked and can't be changed by users
b) using something like the USRP B200
- it can do 61 Million samples/sec, 12 bit samples, 732 Mbit/sec
- but maybe that sample rate is not needed for a band that is 4 MHz wide...
No, 4MS/s should suffice (if you can live with the filter roll-off
at the band edges).
Still, not wasting too much signal quality: for 8bit samples in I
and Q, 4MS/s * 2B/S = 8MB/s = 64Mb/s
c) an instance of GNU Radio taking all the samples and encapsulating
them into packets
d) transmitting to local users
layer 1/physical: 23cm or 13cm, using 8 - 10 MHz bandwidth
So, since AX.25 doesn't specify a modulation (and if we used the
AX.25 that seem to be dominant, we'd end up with a data rate
whopping three to four orders of magnitude too low), let's look at
the data rate here to determine a minimal modulation order and
SNR:
Shannon Channel Capacity says that our bitrate is bound,
if we want to achieve transmission with arbitrarily low bit error
rate over a channel of bandwidth and given
:
I'd say, wow, for a wide-range 10MHz link, that's a pretty good
minimum SNR!
Now, for the modulation:
;
i.e. our modulation would have to have at least that many bits per
symbol, which means at least 85 different states.
Effectively, this calls for something like a 128 QAM, or a 256 QAM
(from a gut feeling, this makes sense if SNR is in fact quite a
bit higher than 19.2 dB) ; more likely the latter, because it's a
square number, making the constellation easier to implement, and
also, because we'll definitely want some bitrate headroom to add
redundancy for channel coding/forward error correction, and , which is pretty handy for code
implementation.
Whether to send those symbols in time-domain or over a set of OFDM
or filterbank carriers would be up for discussion; from an
equalization point of view, using multiple carriers seems to make
a lot of sense; those 10MHz will probably not be nicely flat.
layer 2: AX.25 (with repeater callsign)
As calculated above, not that much room in those 10 MHz for
framing overhead, the relatively ineffective CRC32 and the
5-bit-stuffing, to be honest... I don't think AX.25 is the optimum
choice here. I'd rather go for something that has a usuable
preamble for equalization, and a more compressed header, and
complements the FEC used more nicely.
layer 3: IP multicast (UDP packets)
Why that? If we're going to be fully utilizing the link with
sample packets, anyway, it's not really necessary to have
different logical endpoints, right?
e) Receiving stations would receive the UDP multicast packets and feed
them as input to a flow graph in a local instance of GNU Radio
I can imagine there may be risks with packet loss and the receiving
users may need directional antennas. As it would be a licensed amateur
repeater, it would be able to legally put out more watts than a wifi
router though.
The point is that I lack knowledge about typical SNRs for the 13cm
(2.4GHz) or the 23cm (1.3GHz) bands; problematic for me sounds
that free space loss for 23cm over a distance of 10km would be
around .
So with an minimum SNR of and a thermal
noise floor of , and assuming a relatively
nice receiver with a noise figure :
You'd need a transmit power of
.
So, you can't reliably talk to someone further away than 10km with
a 10kW TX, assuming you have no antenna gains. Sure, a very nicely
aligned dish with low losses can achieve almost 30dB, but that
effectively only means that 1kW is enough for 10km.
Best regards,
Marcus
Regards,
Daniel
|