Cheap SDR devices vs classical communications receivers

[jonpaul]

Main difference is Images, 3rd order IM, front end overload.

A triple conversion superhet is the gold standard

J

[MarkMLl]

My IEEE Spectrum article of 2019 was reconstruction ONLY of one of 72 6 level ADC ("steppers" ) used to quntize the 12 Dudley VOCODER paramentes and true random noise keys used for unbreakable speech scrambing.

See the many artilecs on SIGSALY.

I have a certain appreciation of veteran technology, and was once privileged to work with somebody who'd been at Manchester in the early days of computing. I don't know exactly which machine(s) he worked on but he recounted that he once caused a stir by building a 2Kword memory.

MarkMLl

[A.Z.]

the only cheap ones which are a bit better are those based on Mirics (MSI) chips

Mirics chip is not better, it is worse than RTLSDR. Mirics has a lot of parasitic spurs, dead frequencies, noisy LO and terrible frequency beats. I have both, but almost never use Mirics, due to worse signal quality.

I was referring to dynamic range, which for SDR units is directly related to the number of bits used in the AD

[A.Z.]

My Ettus USRP N200 *always* outperforms the AOR 5000 on the desk next to it -- as long as it has a filter in the line to the antenna.

There is no sense to process 160 MHz bandwidth

The Ettus CAN process up to 160MHz at once, but it's NOT mandatory ... did you ever try one of those ?

[A.Z.]

The newer series of radios from NI/Ettus are truly spectacular. The N320 and the X3xx series are beasts. But they come with a beastly price. (north of $20K) 200 MS/s, full bandwidth back to the computer (twin 10Gb/s links), 14 bit ADC.

Well, if one can accept some limitations but still wants "better than those cheap SDR dongles" performances there are "mid range" products too, like for example

https://winradio.com/home/receivers.htm

https://www.eladit.shop/en-gb/RADIO-SDR-c53003380

sure, those aren't up to par with receivers like the Ettus ones, but are FAR better than the usual "dongles" then it all depends from what one wants to use an SDR unit for

[radiolistener]

I was referring to dynamic range, which for SDR units is directly related to the number of bits used in the AD

No, dynamic range includes ADC bits and bandwidth (ADC sample rate). This is incorrect to compare ADC bits with no take into account working bandwidth.

For example RTLSDR uses 9-bit ADC (SNR=9*6.02+1.76=55.94 dB) running at 28.4 MHz. When you downsample it to 2.7 kHz SSB, the dynamic range will be increased for 10*log(28400/(2*2.7))=37.2 dB. So the dynamic range of RTLSDR is about 55.94+37.2 = 93.4 dB for SSB mode. Which is equals to (93.4-1.76)/6.02=15 bits ENOB.

For comparison Mirics ADC has 12 bit resolution (SNR=12*6.02+1.76=74 dB) at about 250 kHz sample rate. When you downsample it to 2.7 kHz SSB, the dynamic range will be increased for 10*log(250/(2*2.7))=16.6 dB. So the dynamic range of Mirics is about 74+16.6 = 90.6 dB for SSB mode. Which is equals to (90.6-1.76)/6.02=14 bits ENOB.

As you can see, Mirics has a little bit worse dynamic range than RTLSDR, RTLSDR has about 3 dB better dynamic range. 

Of course these calculations don't include some losses on noisy LO implementation (Mirics has much worse LO than RTLSDR), bit rounding and other processing thing. But RTLSDR has better performance than Mirics. The only con of RTLSDR is that it don't allows to get 8 MHz bandwidth IF. But this is not needed for a good receiver, 3.2 MHz of RTLSDR is good enough. Most of the time I'm using 2.4 MHz and it is good enough.

[A.Z.]

For example RTLSDR uses 9-bit ADC (SNR=9*6.02+1.76=55.94 dB) running at 28.4 MHz. When you downsample it to 2.7 kHz SSB, the dynamic range will be increased for 10*log(28400/(2*2.7))=37.2 dB. So the dynamic range of RTLSDR is about 55.94+37.2 = 93.4 dB for SSB mode. Which is equals to (93.4-1.76)/6.02=15 bits ENOB.

For comparison Mirics ADC has 12 bit resolution (SNR=12*6.02+1.76=74 dB) at about 250 kHz sample rate. When you downsample it to 2.7 kHz SSB, the dynamic range will be increased for 10*log(250/(2*2.7))=16.6 dB. So the dynamic range of Mirics is about 74+16.6 = 90.6 dB for SSB mode. Which is equals to (90.6-1.76)/6.02=14 bits ENOB.

If my memory is serving me correctly (and I think it does), the MSI001/MSI2500 sample rate is between 5MHz and 8MHz, not 256KHz (where does that come from ?)

[radiolistener]

If my memory is serving me correctly (and I think it does), the MSI001/MSI2500 sample rate is between 5MHz and 8MHz, not 256KHz (where does that come from ?)

Yes, but it reduces dynamic range for sample rate 5-8 MHz. It's hard to say what exactly mode it uses internally, because this information is not disclosed. But their 12 bit dynamic range is specified for lowest bandwidth which is about 250 kHz. Unfortunately they are showing marketing bits which already include downsample dynamic range gain and there is no information about original ADC specification.

It seems that they hide it because don't want to show real comparison with RTLSDR specification 

And since Mirics using more noisy LO than RTLSDR, it doesn't matter what is dynamic range of their ADC, because dynamic range will be limited with noise of LO.

And after all, Mirics requires to use proprietary driver which is unstable and have many issues with multi-threading and other things. So, this is just a headache to use Mirics, with no any benefits against stable driver for RTLSDR.

For example just compare DC offset of RTLSDR and Mirics. RTLSDR DC offset can be completely filtered with IIR filter with 1 Hz cutoff. Mirics DC offset cannot be filtered even with 100 Hz bandwidth. It shows that Mirics have very dirty and noisy DC offset.

So I just don't understand why people still using it

[A.Z.]

the MSI3101 chipset (MSI001+MSI2500) has 10bit ADC resolution at/above 5MHz rate and can use 12bit resolution below 5MHz rate, at least according to the docs released by Mirics and to the various Linux drivers implementations so, again, I can't see where that "khz" sample rate you referred comes from

[pienari]

SDRplay RSP1A works just fine.

Cheap RTL2832u + 820T2 set i could not receive EFOU_ATIS transmitter 135.450MHz AM 28km from airport.
Rx takes too much interference from FM broadcast transmits 10km away.

With RSP1A EFOU_ATIS 135.450MHz signal is loud and clear S7 to S8.
But what i could not believe with RSP1A i can receive FM radio broadcast from Sweden, 160km to transmitter and signal is S9+30bd with 180khz bw. RDS.

[radiolistener]

the MSI3101 chipset (MSI001+MSI2500) has 10bit ADC resolution at/above 5MHz rate and can use 12bit resolution below 5MHz rate

you're confusing resolution, ADC bit resolution is fixed and specified for specific hardware bandwidth of ADC. It doesn't have sense if bandwidth is not specified. For SDR receiver bit resolution is specified for hardware ADC bandwidth, it should not include processing gain.

There is no such information about Mirics chips in published documentation. They hide this information. The "resolution" you're talking about is different thing, it already includes processing gain. This is just marketing stuff and don't have sense to talk about it if you want to compare it with other devices.

SDRplay RSP1A works just fine.

Cheap RTL2832u + 820T2 set i could not receive EFOU_ATIS transmitter 135.450MHz AM 28km from airport.

Don't confuse ADC performance with presence of input filter. These are different thing. You can get better result with RTLSDR just by adding proper filter on it's input. SDRPlay has such filters inside, but it don't improve it's performancem, it just helps to compensate low dynamic range of SDRPlay ADC by cutting off strong signals.

Good ADC has good dynamic range performance and can work with no issues with presence of strong out-of-band signals on it's input. It doesn't requires additional filters, just because it's dynamic range is high enough to not overload ADC with strong signals at maximum sensitivity.

Adding filters to good ADC just reduce it's sensitivity because filter has some signal loss. So it won't give you advantage, on the contrary - adding filter make signal worse...

If you're still want to use filters with high performance ADC, you're needs to use extra high quality filter with ultra high Q-factor components, proper layout and soldering. Even type of PCB, mask and solder materials can affect results... Such filter will cost more than ADC and cannot provide you with significant improvement on high performance ADC.

[pienari]

Well that is bad if there have to be band pass filter every frequence what to listen.
RTL BW is usable only 2.5MHz vs. RSP1A 10MHz.

I measured RSP1A  NFM sensitivity.
Rx 2.5khz
600kHz 4.3uV snr 10db
1.2MHz 2.1uV snr 10db
3.4MHz 0.95uV snr 10db
6.8MHz 0.78uV snr 10db
10.4MHz 0.8uV snr 10db
14.6MHz 0.6uV snr 10db
18.2MHz 0.6uV snr 10db
22.8MHz 0.6uV snr 10db
25.9MHz 0.55uV snr 10db
29.5MHz 0.55uV snr 10db
47.0MHz 0.55uV snr 10db
56.3MHz 0.62uV snr 10db
61.0MHz 0.27uV snr 10db
102MHz 0.33uV snr 10db
145MHz 0.33uV snr 10db
205MHz 0.30uV snr 10db
435MHz 0.24uV snr 10db
835MHz 0.21uV snr 10db
1040MHz 0.24uV snr 10db

1040 is top of my signal generator.

[radiolistener]

RTL BW is usable only 2.5MHz vs. RSP1A 10MHz.

That is usual misunderstanding.

As I said before, RTLSDR ADC working at 28.8 MHz, it means that it's ADC has bandwidth 14.4 MHz. This is more than 10 MHz

2.4 MHz is IF bandwidth after processing. Actually RTLSDR allows IF up to 3.2 MHz.
There is no sense to have so wide IF bandwidth. 1-2 MHz is good enough.

145MHz 0.33uV snr 10db

According to my tests RTLSDR has much better sensitivity on 145 MHz band, it can receive very weak signals when Mirics is already flooded with noises.

At HF band they have the same sensitivity, but Mirics has many dead frequencies and more spurs. RTLSDR has mirrors around 14.4 MHz, but it can be fixed with filter or external up converter. This issue is already fixed in RTLSDRv4.

[pienari]

I have old AOR 8200 mk2 and i have compared it to SDR . I don't need AOR anymore , SDR outperforms

[radiolistener]

I measured RSP1A  NFM sensitivity.

I think this is a bad idea to measure sensitivity with FM modulation, because it has non linear properties, so you can get strange results. It's better to use SSB or CW for that. Usually sensitivity is measured with 2.7 kHz bandwidth for SSB and 500 Hz for CW.

[fourfathom]

Sometimes you just can't use a narrow front-end filter.  I had mentioned the RX-888 and external shelf / anti-alias filters up-thread (https://www.eevblog.com/forum/rf-microwave/cheap-sdr-devices-vs-classical-communications-receivers/msg5106288/#msg5106288), used in full-spectrum reception.  I know several people who are running this setup during today's eclipse, for propagation studies.  The RX-888 is receiving the full 0-30 MHz spectrum (the only front-end filter is the shelf/LPF), at a sample rate of around 66 MHz and sending about 128 MBytes per second over the fast USB connection to the computer.  Using a 3.5 TByte drive they can record six hours of full-spectrum data for later analysis.

[radiolistener]

at a sample rate of around 66 MHz and sending about 128 MBytes per second over the fast USB connection to the computer.  Using a 3.5 TByte drive they can record six hours of full-spectrum data for later analysis.

omg, this is a waste of disk space 
but if I had a 3.5 TB disk I would try 

[pienari]

I measured RSP1A  NFM sensitivity.

I think this is a bad idea to measure sensitivity with FM modulation, because it has non linear properties, so you can get strange results. It's better to use SSB or CW for that. Usually sensitivity is measured with 2.7 kHz bandwidth for SSB and 500 Hz for CW.

Well .. if you have such devices, would you please make some measurements for us?

[A.Z.]

I measured RSP1A  NFM sensitivity.

I think this is a bad idea to measure sensitivity with FM modulation, because it has non linear properties, so you can get strange results. It's better to use SSB or CW for that. Usually sensitivity is measured with 2.7 kHz bandwidth for SSB and 500 Hz for CW.

Well .. if you have such devices, would you please make some measurements for us?

that would be really interesting, also since the results, if unbiased, could reveal a lot of things

[radiolistener]

Well .. if you have such devices, would you please make some measurements for us?

Yes, I can do, but I don't have good signal source, just Chinese PSG9080 and Si5351 and ADF4351. I have attenuators, so can make any signal level, but the PSG9080 output is not so clean and Si5351 / ADF4351 has square wave which may affect results for some tests.
What kind of measurement do you want to see?