Characterizing the RTL-SDR Block (Version 3) for HF reception using the "direct" input.

An inexpensive option to reception on the HF (low frequency) bands is one sold by "RTL-SDR Blog" - the current iteration being Version 3.  Originally intended for digital VHF/UHF TV reception - and that of FM broadcast - the hardware is also capable of tuning much lower frequencies.

Figure 1:  An RTL-SDR Blog V3.

Unlike most other inexpensive RTL-SDR dongles, this has - via a single SMA port - the ability to operate in "direct" mode where RF is passed straight to the A/D converter rather than via a down-converter, allowing reception from (theoretically) a few hundred kHz to around 30 MHz.

There's a penalty to pay for simplicity:

For those familiar with receiver topology - digital or analog - several things the above description of the HF operation of the RTL-SDR dongle without proper measures (e.g. filtering, gain control) should give cause for concern - namely:

• There is NO bandpass filtering at all.  Whatever is being intercepted by the antenna system will be input directly to the A/D converter.  
  

• ANY signal applied to the antenna input - no matter the frequency - can contribute to overload.  Because the inputted RF goes into the A/D converter, a strong signals well away from where you are listening can cause overload.  For example, if you are listening to 14 MHz, a strong AM broadcast signal (e.g. mediumwave) could well be the culprit if you are experiencing overload.
  

• There is no AGC in the HF ("direct") signal path.  Considering that the A/D converter is only 8 bits, this means that compared to even low-end shortwave receivers, the range of signal levels over which this device will operate is very narrow.
  

• The A/D converter's sample rate is 28.8 MHz.  What this means is that the Nyquist limit - the frequency above which the digitized output can no longer represent the input signal is half this, or 14.4 MHz, uncomfortably close to the 20 meter amateur band, and entirely below the 17, 15, 12 and 10 meter bands - and this does not even consider the fact that the sample frequency is within the 10 meter band itself.  What this means is that there are unsuppressed image responses, the frequency of which is calculated by subtracting it from the sample rate (e.g. 28.8 MHz).  For example, 15 meter signals will also appear, spectrally inverted (e.g. USB = LSB) 7.80-7.35 MHz.  The image frequencies appear in the table - along with MDS and clipping values - below.

In other words, if you are tuned to any HF frequency, you are actually "hearing"TWO frequencies at the same time unless you have specific filtering to prevent such image response.  See the right-hand column of Table 1 (below) for the frequency at which you will see an image.

The above put strict limits on the performance of the RTL-SDR dongle as an HF receiver and these realities must be considered when the configuring any system that might use them.

The usable dynamic range:

In theory, 8 bits of A/D sampling would indicate about 48dB of useful signal range, but the reality is more complicated than this.  Compared to the bandwidth of the narrow signals typically sought on HF, the overall sample rate of the A/D converter - and even the rate after the RTL2832U's converter has reduced the signal to the sample rate being sent to the USB port - are very much higher, effectively improving the bit depth via oversampling - and the fact that the HF spectrum is backgrounded with what amounts to white noise works to our benefit, helping to spread discrete, spectral artifacts that are the inevitable result of the imperfect signal aquisition.

All of this makes the actual, usable range a bit difficult to divine.  To this end, an RTL-SDR (V3) unit was put on the workbench, using the "HDSDR" program as a receiver, and its operation was analyzed by observing CW (unmodulated) signals on narrow (SSB) bandwidths.

Methodology:

Ideally, we would be able to determine the RF level at which the A/D convert clipped directly, but the HDSDR program does not provide a means to see the peak A/D level, requiring us to infer it by noting the level from a known-accurate signal generator at which the S-meter starts to decrease at the same rate that the signal level is reduced - and also by noting the disappearance from the waterfall many of spurious signal caused by overload.  Typically, this is about 3dB below the "maximum" S-meter reading.

For sensitivity, DL4YHF's "Spectrum Lab" program was used to measure the SINAD in a 500 Hz bandwidth set by the HDSDR, the "minimum discernible signal" being equivalent, in this test, to 3dB S/N.

Discussion:

Table 1 tells us several things:

1. As mentioned in the documentation found at the RTL-SDR Blog web site, signals below 2 MHz - and especially below 1 MHz - are rolled off by the "Bias Tee" blocking choke which has insufficient inductance at frequencies below the AM broadcast band.  To a degree, this effect can be mitigated by removal of the RF choke which will remove the capability to inject DC onto the cable.
2. The RTL-SDR has a low-pass filter to diplex the HF and VHF and above frequencies to separate signal paths, and the effects of this filter are becoming evident above 15 meters (21 MHz).  This also means that by itself, the RTL-SDR becomes "deaf as a post" on the 12 and 10 meter bands.
   
3. The effects of high-pass roll-off appears to be evident through at least 7 MHz.
4. There is clearly no image rejection at all, considering that the sensitivity through 21 MHz is comparable to that below 14 MHz.  For example, 15 MHz WWV will also appear at 13.8 MHz.
   
5. Low-pass effects are evident by 24 MHz, most likely a result of the diplexer used to split the HF (direct) and VHF/UHF signal paths.  This limits sensitivity on the 12 and 10 meter amateur bands.
6. With only 8 bits of quantization, additional noise will be generated due to the imprecise nature of the process.  This "noise" will show up as spurious signals and, less obviously, as a rise in the overall noise floor - depending on the nature of what is being digitized.  In short, the fewer the number of bits, the less likely it is that weak signals will coexist (and be audible) in the presence of strong signals.
7. Across most of the HF spectrum, the RTL-SDR will overload signal of about -30dBm - which is approximately equal to an S-meter reading of "40 over S-9".  While this seems like a fairly strong signal, this power level represents the total amount of RF energy - no matter the frequency!
8. Note that the power at which the A/D converter starts to overload is approximately -30dBm across much of the HF spectrum.  This represents the TOTAL amount of RF power required, at all frequencies combined, that will result in overload.

Points 7 and 8, above, should be considered very carefully in terms of its implications:

• You cannot simply connect an RTL-SDR Dongle to even an "average" performing HF antenna and expect reasonable results as the total power from ALL signals reaching the receiver are likely to exceed the -30dBm signal level.
  

• Particularly on the lower bands (80, 40, 30 meters) the signal levels of amateur and especially shortwave broadcast signals can, by themselves, exceed the -30dBm overload level - particularly in Europe and the eastern U.S. On some bands, the shortwave broadcast band adjacent to the amateur band (e.g. 41 and 40 meters) are too close to effectively filter out and it may be that an RTL-SDR is simply not usable on these bands when propagation favors reception on these frequencies. 
  

• As seen in Table 1 (above) the RTL-SDR dongle becomes increasingly deaf on the higher HF bands (particularly 12 and 10 meters) making them unusable at these frequencies without additional amplification AND filtering.
  

• If you plan to use an RTL-SDR for HF reception and expect even mediocre performance, you should precede it with a band-pass filter for the frequency band of interest.
  

• It's worth remembering that unlike an analog receive system, one cannot always use all 8 bits for digitization:  The signal input must be kept well below the "full scale" level to accommodate for random fluctuations that are ever-present on signals input from the antenna.  What this means is that the A/D must be under-driven overall and that fewer bits are actually being used most of the time.  In order to maintain suitable margin, it's typical to drive the A/D converter at between 1/4 and 1/2 full scale, meaning that for 8 bits of A/D conversion, 2-4 bits are typically being used.
  

Conclusions:

• DO NOT simply connect an RTL-SDR to your HF antenna and expect it to work as well as even a low-end shortwave receiver:  If it doesn't get overloaded by local AM broadcast (mediumwave) signals, it will get overloaded by strong shortwave broadcast signals when conditions are favorable on certain bands.  This isn't to say that you won't hear anything if you do so, but know that normal signal levels present on even an "average" antenna will be enough to overload the RTL-SDR dongle.
  

• ALWAYS precede an RTL-SDR with a band-pass filter that is specific for the frequency range of interest. For example, if you are interested in 40 meter reception (7.0-7.3 MHz for ITU region 2) your filter should pass only frequencies in this range, and this filtering will prevent unwanted reception of signals at the image frequency around 21.8 MHz.   Unfortunately, the use of a band-pass filter precludes reception outside its design range, but this is necessary considering the limited capability of the RTL-SDR dongle in terms of handling both strong and weak signals at the same time and its unfettered response to unwanted images.
  

• In some cases - even with a mediocre antenna - the signals in the desired frequency range may exceed the signal handling capability of the RTL-SDR and cause overload.  As noted above, a single signal stronger than -30dBm can do this, but so could a number of signals whose total power can exceed this.  Typically, this overload is manifest as intermittent distortion across the entire receiver as signals fade in and out.  In such cases, it might be beneficial to attenuate the signals reaching the RTL-SDR as the degradation caused by overload is more "destructive" across the entire receive frequency range than too-little signal.
  

• Because of image response and roll-off, the HF port of the RTL-SDR is really not well-suited for 12 and 10 meter (24-30 MHz) reception.

Comment about HF upconverters:

Specifically to address some of these issues, there are upconverters available for the RTL-SDR that will upconvert the HF spectrum to VHF (typically in the 100-130 MHz range) to allow the use of that signal path, making use of the Raphael R820T converter.  This converter has the advantage of having a degree of band-pass filtering and the ability to use AGC (automatic gain control) on the signal path.

This method can be useful, but there are several caveats:

• If frequency stability is of importance, the addition of the upconverter introduces two additional frequency stability issues:  Drift of the upconverter itself, and the fact that any existing drift in the dongle itself will be multiplied because of its operation at the higher frequency.  This can be an issue in environments where the temperature is not stable and/or when a frequency sensitive mode like SSB or (especially) digital modes are used.
  

• If multiple bands and receivers are to be used, there may be the temptation to upconvert the output of the upconverter to VHF and distribute this signal to the receivers.  While this may work in many cases, it's worth noting that if the entire HF spectrum is converted, the total signal power level can be significant, potentially overloading the upconverter itself, any RF amplifiers that might be used at VHF, and/or the front end of the RTL-SDR dongles themselves.
• While a "fix" might normally be to filter out just the HF band(s) of interest, this can become impractical if the 40 meter band (7.0-7.3 MHz) is upconverted to, say, 137 MHz where it can become difficult to make an effective band-pass filter at that frequency.
• It is possible to filter for a specific HF band before the upconversion, but this means that each RTL-SDR would require its very own upconverter.  While effective, the cost of an upconverter for each band may be prohibitive.

* * *