Skip to content
forked from keenerd/rtl-sdr

RTL-SDR *very* experimental branch - it's probably broken!

License

Notifications You must be signed in to change notification settings

mutability/rtl-sdr

 
 

Repository files navigation

rtl-sdr
turns your Realtek RTL2832 based DVB dongle into a SDR receiver
======================================================================

For more information see:
http://sdr.osmocom.org/trac/wiki/rtl-sdr


Notes on the mutability tree (2014/09/29)
=========================================

Since there seems to be some wider interest in this, some quick notes on
my changes here.

TL;DR:

0Hz
 .
 .  "no mod direct sampling"
 .  no mixing at all, tuner is mostly disabled, just ram as much signal
 .  as you can at it and maybe some will leak through.
 .  YMMV a lot
 .
~13MHz  (PLL lower limit - 14MHz)
 .
 .  normal tuning, large IF.
 .  high-side mixing
 .  you will see nasty aliasing / attentuation / harmonics
 .  near the lower edge of the range
 .  you will see lots of noise from the dongle itself near 14.4MHz
 .
~21MHz  (PLL lower limit - 6MHz)   <- upstream tuner lower limit
 .
 .  normal tuning, regular IF
 .  high-side mixing
 .  this should behave much like upstream
 .
~1844Mhz (PLL upper limit - 6MHz)  <- upstream tuner upper limit
 .
 .  normal tuning, small IF (getting squashed against the PLL upper bound)
 .  high-side mixing
 .
~1848MHz (PLL upper limit - 2MHz)
 .
 .  tuning with low-side mixing (PLL frequency below tuned frequency)
 .  you will see nasty aliasing / attentuation / harmonics
 .  near the upper edge of the range
 .
~1864MHz (PLL upper limit + 14MHz) <- that's all, folks

(PLL limits vary by dongle - some go as high as 1885MHz)

This tree is a collection of random tuner hacks which are really exploratory
more than anything. They may or may not work for you.

First some background for those unfamiliar with the internals of the dongle:

 * The R820T tuner has a tunable PLL that can generate frequencies between
   around 27MHz .. 1850MHz. The exact range varies from dongle to dongle.
 * The tuner mixes the incoming RF signal with this PLL output. This shifts
   the RF signal down to a frequency that is the difference between the PLL
   frequency and the RF signal's frequency.
 * The tuner sends this intermediate frequency (IF) signal to the RTL2838U
 * The RTL2838U digitizes the IF signal and does a digital downconversion
   step to generate I/Q samples that are centered around "zero". The
   downconverter can theoretically handle IF signals up to 14.4MHz.

The main change is to feed information about the actually-tuned frequency back
from the tuner (R820T) code to the core (RTL2832U) code, allowing the core
code to adjust to the actual IF in use, rather than requiring a fixed IF. The
core code has also been taught how to handle low-side mixing, where the PLL
frequency is lower than the requested frequency (e.g. the spectrum inversion
changes then)

When tuning, the R820T tuner will try to tune to the requested frequency +
6MHz, producing a 6MHz IF. If the PLL cannot handle that frequency, the tuner
will try a few things to produce something useful.

At the top of the tuning range:

 * It will tune as high as it can go normally, then stick there producing a
   smaller effective IF as the requested frequency gets higher and higher.
 * Once the IF starts getting too small (below about 1.5MHz things start to
   go bad), it will switch to low-side mixing and try to put the PLL frequency
   *below* the target frequency. This gives you another 14-15 MHz at the top
   of the range.

I've had reports that some dongles can tune to within a whisker of 2000MHz
with this.

At the bottom of the range:

 * It will tune as low as it can go normally, then stick there producing a
   larger effective IF as the requested frequency gets lower and lower.
 * Once the required IF exceeds 14.4MHz, it will switch to a variant of tejeez'
   et al "no mod direct sampling" mode. This essentially disables the PLL in
   the tuner entirely, and relies on some RF signal leaking through the tuner
   unchanged. The tuner claims to be tuned to 0Hz in this mode, and the core
   does all the real tuning. The dongle is almost deaf in this mode, you will
   have to turn the RF gain WAY UP. You do not need to mess with direct sampling
   settings to enable this, it's integrated with normal tuning.
   
The success of the "no mod direct sampling" mode seems to vary a lot from dongle
to dongle. Mine is almost completely deaf so I can't test this much. Others have
had success in tuning to shortwave stations as low as 3.5MHz.
   
The actual PLL limits vary from dongle to dongle, and from day to day (they're
probably temperature related). The tuner has three sets of tuning limits:

 * a hardcoded "do not exceed this" set of limits - see PLL_INITIAL_LOW /
   PLL_INITIAL_HIGH in tuner_r82xx.c. These are in place because, especially at
   the low end of the range, the PLL can get into a state where you ask for (for
   example) 25MHz, the PLL claims to be locked OK at that frequency, but in
   reality it's actually producing 27MHz, which screws up the core's calculations
   of the IF offset needed.
 * a hardcoded "known to be OK" set of limits - see PLL_SAFE_LOW / PLL_SAFE_HIGH
   in tuner_r82xx.c. This is a range where the PLL should always work; if the PLL
   doesn't work in this range it is treated as a tuning error and tuning fails.
 * a runtime "seems to be OK at the moment" set of limits. This varies from run
   to run and initially starts at the "do not exceed" limits. Whenever a failure
   to get PLL lock is seen, the runtime limits are shrunk accordingly and we try
   again. This allows the tuner to adapt to the particular dongle in use.
   
Remember that at the extreme edges of these ranges you will be fighting an uphill
battle against the dongle design itself, as you will be far out of the intended
range of the filters etc and nothing works particularly well.

You are likely to see small frequency shifts (maybe 1-2ppm) from what you get in
upstream. This is because the tuner+2832 combination can be tuned a little more
accurately than upstream does, so some errors that you will have previously been
correcting for will disappear.


-- Oliver ([email protected] / [email protected])

About

RTL-SDR *very* experimental branch - it's probably broken!

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • C 97.2%
  • Shell 1.5%
  • C++ 1.3%