Addition of 100nF, 1nF and 100pF bypass capacitors on the power supply rail.
Added a common mode choke to the 5V line.
Added a MuRata NFM21 EMI suppression filter to the 5V line.
Replaced the oscillator with a 0.3 ppm temperature controlled oscillator (TCXO).
Disabled the internal RTL2832U 1.2V switching supply and provided external 3.3V and 1.2V supplies.
Replaced the MCX connector with an SMA female connector.
Enclosed circuit in a metal box.
In addition to the mods, Laidukas also made some measurements on the performance of the R820T2 on some metrics. In the first test he measured the input insertion loss or SWR. He found that the SWR was below 2 between frequencies of 25 MHz to 1076 MHz. At higher frequencies the SWR reached levels up to about 8.
Another test showed that with the LNA disabled the R820T2 had a lower noise floor by about 7dB, when compared to the R820T.
However, he has continued to modify the RTL-SDR using the same principal (Japanese, use Google Translate). He discovered that to fully suppress the 1.2v switching regulator pin 27 on the RTL2832U chip must be connected to ground. Once this is done the switching noise present at 1 MHz disappears, although there does not seem to be any significant improvement in noise anywhere else.
The first screenshot shows an unmodded dongle with the switching regulator still connected. There is some noise present at around 1.026 MHz. The second image shows a modded dongle, the noise at 1.026 MHz is gone.
By using an upconverter, direct sampling mod, or experimental software driver the RTL-SDR can be used for HF reception. However, a problem with HF reception and the RTL2832U chip is that it uses a switching power supply in its design. This switching power supply causes significant amounts of noise spurs to appear in the HF spectrum.
Japanese RTL-SDR experimented ttrftech has recently posted about a modification to the RTL-SDR which he performed (note in Japanese, use Google translate to read). The purpose of the switching power supply in the RTL2832U is to step 3.3v down to 1.2v. In his modification, ttrftech disconnected the switching power supply and instead stepped 3.3v down to 1.2v by using three diodes. This works as each diode has a voltage drop across it of 0.7v.
His results show that there is a significant reduction in noise spurs at HF frequencies.