Direct Digital Sampled RF Transceiver

[Saber]

Anyone here ever tried something like this?

This is a project in the making, that has been going for a while now (mostly conceptual/planning/reality checks up till now).  This is possible with current technology, but bloody expensive to pull off.....that is if you are "allowed" to buy the components (ever tried to buy radiation hardened FPGAs before?).

My initial requirements were:
- HF right up to UHF (0 - 500MHz) - all HAM modulations (including squeezing Codec2 into there)
- No decimation/mixing in hardware at all
- DSP for everything between the ADC and DAC (16-bit)
- Initial implementation HF only, but solution had to be able to go to a 500MHz modulated signal.
- and lastly, a simple digital LCD screen/touch/knobs/speaker solution to fill in the rest of the radio (preferably classy - technology without aesthetics is just not on).

The whole idea behind this is to filter and amplify your RF signal, and then directly sample the RF signal via the ADC, and everything leaving the DSP would exit through a DAC. All further processing (IF mixing, filtering, modulation, demodulation) is done digitally. Although it might look like a pointless exercise, it can be done....and I think one reason might be to separate the RF standard(s) from the hardware. If anything changes, you just update your firmware. Another reason would be to start implementing the open source Codec2 into hardware (I'm a big fan)

Now......I know it can be done in ways that are a lot easier and cheaper....not the point.  I want to do it this way. My problem is running into FPGA/DSP limits for the required 10Gbps LVDS serial Iine I require for the sampled data. My initial though process went down the route of DSPs with high frequency ADC/DACs, but after eventually locating the DSPs that had the oomph to do what I needed it to do.....I discovered better ADC/DACs, and the DSPs cannot keep up anymore (the story of my life....starts to look like my 25 year old fiber home automation system - don't ask)....and the Freescale DSPs are friggin expensive.

I have now (sulking) scaled back the initial concept. I can use cheaper ADC/DACs, but I'll have to stagger them to stay within the performance limits, and I've decided that FPGA is a better route to go (the FPGA would also be better at regulating the staggered ADC/DAC clocks with minimal jitter). I am now also prepared to decimate in hardware to get the I/Q signals (sulk).

Have anyone here ever done something similar to this? Experimental/commercial/professional, I don't care. If you have any guidance, input, or ideas as far as is possible on component selection, possible architectures. I'll appreciate it.

This was hard work. Opening a can of Windhoek draught now........ (what?, no icon for beer? this is just not on in the engineering world) 

Regards

Johan
 

[PA0PBZ]

Did you ever google SDR?
The common way to do this is ADC and FPGA as a frontend, then send IQ to a PC and do baseband processing there.
Decimation in the FPGA, for most stuff you really don't need the full bandwidth.
As an example look at the block diagram of the QS1R.

[uncle_bob]

Hi

What you have described is essentially what is in a < $20 USB SDR dongle. They have simply smashed it down into ASIC's rather than stopping at the FPGA level.

That said, sure it work. Conceptually it is the way about 99% of all the receivers are done these days. (They make a *lot* of cell phones). You do it with different parts depending on your production volume.

Bob

[Saber]

The point is to do this as a standalone transceiver. Just like a normal HAM base station, just digital all the way. Most of us HAMs (and techies) like our switches, and buttons, and lights, and switches, and buttons, and lights......the more lights the better.

I do understand that functionally, everything I want to do exists, but there is still something to be said for sitting in front on a device with buttons on it........like a real scope, or speccie, or gennie (wondered why the USB scope never caught on). .....and whether we like it or not, the analogue HAM station is on it's way out.....we need a replacement.

Johan

[uncle_bob]

Hi

People do make radios like you describe. Regulatory limits restrict what they can do transmit range wise. Last time I looked, you could do pretty well on a full range of knobs / buttons / dials / lights and have digital inside for under $10K. For the "plugs into the computer and displays there" the price is quite a bit less.

The why on the price mostly relates to volume of manufacture and tooling. You don't see a lot of them because people don't thunk down that kind of money on a single toy very often.

Bob

[Saber]

Uncle_bob.....KACHING.....

If we wait for Yaesu, Icom, Kenwood, we are going to get overpriced radios (again), that is if they are prepared to continue manufacturing radios for HAMs in the future (small market).

$10k in ZA accounts for a few years of income for some people (and I'm not talking about squatters).

FFS.....we are all mostly engineers, aren't we? David Rowe (PhD in audio coding) created one of the most efficient, low bandwidth, real-time, codecs available today....open-source. Let's continue the trend and create hardware to match....

Johan

[Saber]

...btw...you cannot go for regulatory compliance until you actually have something....very important to consider, but not a problem on day one.

Johan

[uncle_bob]

Uncle_bob.....KACHING.....

If we wait for Yaesu, Icom, Kenwood, we are going to get overpriced radios (again), that is if they are prepared to continue manufacturing radios for HAMs in the future (small market).

$10k in ZA accounts for a few years of income for some people (and I'm not talking about squatters).

FFS.....we are all mostly engineers, aren't we? David Rowe (PhD in audio coding) created one of the most efficient, low bandwidth, real-time, codecs available today....open-source. Let's continue the trend and create hardware to match....

Johan

Hi

I've run electronics business over the years in addition to being an engineer. There is no way you are going to come up with a product to compete with the Japanese at a lower price. Your volumes will be lower and your tooling costs will eat you alive. For what they do, the Japanese radios are dirt cheap. In some cases the top of the line radio sells at a loss. It's just there for the prestige factor.

Bob

[Saber]

Not trying to compete with anyone, or start a business. Ever heard of the love of a hobby?

Anyway, if you have nothing to contribute (positive or negative - and you provided neither), I'd prefer to hear from the real engineers.

Johan

[donmr]

That is NOT a $20 SDR, those mix the signal down first then sample it.

Look at the HPSDR project, they did it for up to 60 MHz.

To get 500 MHz you will need 1GHz A/D(s), tha twill cost a lot.

[AF6LJ]

These folks have been doing what you are talking about however not to the extent of making it to 500MHZ that is going to be a bit much...
http://www.flexradio.com/

[uncle_bob]

Not trying to compete with anyone, or start a business. Ever heard of the love of a hobby?

Anyway, if you have nothing to contribute (positive or negative - and you provided neither), I'd prefer to hear from the real engineers.

Johan

Hi

Here's the problem with that. When you shut out anything you "don't like" it's not much of a conversation.

Indeed I have designed and built these sort of radios and do very similar stuff of a living. It that's not a qualification for being in the conversation ... it's a pretty exclusive club.

Bob

[ConKbot]

Selectivity is what makes good receivers good.  How would you handle a ham transmitting at few hundred watts on one band when you're trying to pull something in down in the dirt on another band? I.e. A random yaesu unit I looked at had a sensitivity of 0.2uV or -120dBm. If you dabble in a few bands, and have an antenna array, and your neighbour ham is putting -30 dBm into your antenna, you're eating up 90db of dynamic range before you even get to your signal.
You're looking at a 18+bit ADC to even consider that.  A few hundred dollars just into the ADC.
At least with the SDRs it's bandlimited to 60MHz or less (depending on model), so once you're into the shorter bands, there is some selectivity.

[uncle_bob]

Selectivity is what makes good receivers good.  How would you handle a ham transmitting at few hundred watts on one band when you're trying to pull something in down in the dirt on another band? I.e. A random yaesu unit I looked at had a sensitivity of 0.2uV or -120dBm. If you dabble in a few bands, and have an antenna array, and your neighbour ham is putting -30 dBm into your antenna, you're eating up 90db of dynamic range before you even get to your signal.
You're looking at a 18+bit ADC to even consider that.  A few hundred dollars just into the ADC.
At least with the SDRs it's bandlimited to 60MHz or less (depending on model), so once you're into the shorter bands, there is some selectivity.

Hi

Even more practically, you have the basic issue of keeping your transmit power all in one band. It may be a hobby, but there are rules

The same (likely band pass) filters that take care of the transmit can help you with the receive. You still have the problem with Bob next door running a KW 10 KHz up the band. That's a problem with any radio out there.  Some sort of filtering and sub-sampling might also keep you from needing to go shopping for 16 to 18 bit 1.25 GS/S ADC's with 120 db SFDR.  (500 MHz plus Nyquist filter gets you to above 1 GS/S otherwise). You do get *some* benefit from decimation so exactly how many bits .... TBD. Dropping the rate might also impact your FPGA cost quite a bit as well.

Bob

[AF6LJ]

This is an interesting discussion; I once heard a talk at a local hamfest by a gentleman who was well qualified to discuss the subject we are here. One of the biggest limiting factors of direct sampling, or for that matter just about all the current crop of SDR radios is receiver dynamic range. There is all kinds of software slight of hand that can be used to reduct the spurious emissions to an acceptable level but with even a 16 bit A/D converter dynamic range is limited to levels that are well below state of the art receivers and even many from years past.
http://www.sherweng.com/table.html

[cdev]

Are you sure?

Sometimes, the more direct perhaps the better, when the alternative is shift signals to some IF from every which where - at least in solid state equipment that seems like a process prone to creating spurs.

There is no comparison between a good SDR and most conventional radios..

8 bit SDRs though, I totally agree... With waterfall displays sometimes a screenshot will tell you more than an audio report about the ability to receive a wide dynamic range.  the cleanest screenshots I have gotten with a DVB-T dongle have been direct sampling.

But the gain is hard to get right. Its never quite right, you have to ride the gain a lot. Also, use attenuation, preselectors, etc.

[uncle_bob]

Hi

A 1970's vintage high performance radio will have filters in it with -135 db ultimate selectivity. The synthesis chain in that same radio will be spur free to a similar level. That level of performance simply got better as time went on. Commercial grade VHF radios in the 70's had similar specs (they stated them a bit differently).

An ADC (let alone a > 1 giga sample part) with a SFDR past 100 db is not a common item:

http://www.analog.com/en/products/analog-to-digital-converters/high-speed-ad-10msps/high-speed-ad-10msps.html

Sort by sample rate first and then look at SFDR. The first two parts are the only ones that give you a chance at 500 MHz. The faster one is 12 bits, the slower 14 bits. The 14 bit part has a 77 db SFDR. You can run through the whole > 10 MHz table and not spot anything with a listed SFDR past 100 db.

Decimation tricks and processing gain *will* help your noise issues. They will also add a bunch of bits. None of that takes care of SFDR. You can only fool mother nature just so far.

Now, how much range *do* you need? Sone designs seem to be ok at 60 db. If the objective is a super radio that beats anything on the planet, the spec's are a bit tougher. Out in nowhere on an empty band is not as much of a challenge as 10 guys in town all going power crazy.

Bob

[cdev]

Here we go..  more bits, less bucks.. Just get a better sound card.. You may already have one.

QSD is inherently also a good band pass filter!  This is what Flexradio and Softrock and a lot of other quite economical quite good SDRs use. It really works well. Its like a rotary switch, click click click click.

http://www.google.com/search?q=quadrature+sampling+detector

http://www.google.com/search?q=tayloe+mixer

So your sound card is your ADC. And thats around the cheapest high quality ADC you are going to find anywhere and it works in the low frequency domain so the QSD converts your signals to that.  I and Q.

Ideally, you then get a 24 bit sound card that can sample at 192 KHz..

 24 bit sampling is substantially better than 16 bit. the difference is really quite dramatic. Its much much better than a 55 or 60 db dynamic range in an RTLSDR, it seems to me. 

Absolutely no comparison. With an RTLSDR you almost never hear the weak stuff at all.

You want to maximize the voltage the sound card input can handle - so that it can handle the full voltage that the receiver can output without clipping. The floor is whatever the input noise reads when they are shorted.

Also, you should use offset tuning so you are never smack dab on the middle.

[uncle_bob]

Hi

If you have a sound card as the digital part, the best you are going to do for bandwidth is ~ 200 KHz. To do that without weird stuff happening, you need stuff like mixers (samplers) and a variable clock (synthesizer). You also will need something ahead of the sound card to take care of Mr Nyquist. (= selectivity)

The practical real world solution is to sacrifice SFDR for simplicity and lower cost. The radio isn't quite as good, but it is the best one you can build. It has other areas where it blows the old analog beast out of the arena. It is only when you say that you want the best of all things all at the same time (possibly at a real low price as well) that the process goes "tilt".

Bob

[cdev]

Hi

If you have a sound card as the digital part, the best you are going to do for bandwidth is ~ 200 KHz. To do that without weird stuff happening, you need stuff like mixers (samplers) and a variable clock (synthesizer). You also will need something ahead of the sound card to take care of Mr Nyquist. (= selectivity)

Yes, youre right.

Not even 192 KHz, with your 192 KHz sound card, you will get a solid maybe 180 KHz bandwidth with rolloff at the edges.

The practical real world solution is to sacrifice SFDR for simplicity and lower cost. The radio isn't quite as good, but it is the best one you can build. It has other areas where it blows the old analog beast out of the arena. It is only when you say that you want the best of all things all at the same time (possibly at a real low price as well) that the process goes "tilt".

I can live with those tradeoffs.

[cdev]

Ive never seen off the shelf (consumer) sound cards with more than a 192 KHz - (although I would not be surprised if they were out there, I just have never seen or used one)  and it seems that since the human ear only goes up to at the most 20 KHz, its unusual that they care what the passband is above that but still, many of them work okay, my sound card is a realtek onboard sound card and as I said I get a solid 170-180 KHz or so of BW with rolloff at the edges.

they do have filters for nyquist/antialiasing built in..or so I have read.

If sound cards were made for SDR use, they would be a lot more expensive because lower volume.

[Kalvin]

Unless you are using really wide modulation bandwidth there is not real reason or benefit to use direct RF sampling ADC and direct RF modulation with a DAC. Your sensitivity will be very poor if you are using a wideband receiver for receiving relative narrow band modulation.

Mixing to a relative narrow IF band and sampling this IF with an ADC, or using a zero IF architecture, will get the sensitivity as you can amplify the frequency in interest in order to maximize the dynamic range of your ADC. In similar fashion a transmitter can be created using a DAC to output relative narrow band IF signal which is then mixed to appropriate frequency.

Commutating detector ie. Tayloe detector is popular direct conversion architecture in SDR community. The commutating direct conversion architecture can also be reversed and used as a transmitter. Due to the practical component limitations, the commutating technique is usable up to 100MHz or so as the analog switches used in the detector need to switched at rate of 4 x frequency of interest. For example, listening for a FM radio at 100MHz, the switching frequency for the analog switches needs to be 400MHz. At higher switching frequencies the detected signal quality start suffering from the imperfections of the analog switches.

Using a transverter scheme, the direct conversion and direct IF sampling/modulation techniques can be used to create versatile and well performing transceivers.

[dkozel]

Hi Saber and folks.

Right out front, I work for Ettus Research and really like SDR.

Here's an example of an Amateur Radio operator who decided to do very similar to what you want to do. Ettus Research started off as a project to design a low cost, extensible software defined radio. The first radio was the USRP1 which was a motherboard with USB, an ADC/DAC, and an FPGA and a series of daughterboards for different frequency ranges with amplifiers, filters, and mixers.

The schematics are online here as well as two descriptions of the radio.

• http://files.ettus.com/schematics/usrp1/usrp1.pdf
• http://files.ettus.com/manual/page_usrp1.html
• http://www.faculty.ece.vt.edu/swe/chamrad/crdocs/CRTM09_060727_USRP.pdf

All the source code of the firmware, driver, and FPGA image are also available under the GPL license.

• https://github.com/ettusresearch/uhd

The USRP1 of course is basically entirely outdated at this point. There have been a bunch of radios released since then, the snazziest one receiving 120 MHz each from two daughterboards, tunable 10 MHz to 6 GHz. The schematics and source code are available at the links above.

Hopefully you don't take this as a sales pitch, just some resources for seeing what one group of people's designs look like for a broadband SDR. Let me know if you have any questions. GNU Radio is good fun to play with as well for experimenting with DSP.

There's a lot you can do with a broadband SDR without needing too much filtering, but it does certainly help! I have a stack of filters around my apartment which I swap in depending on my signal of interest. It helps deal with local FM broadcast, wifi interference, and my regular ham radio transmissions. Also knocks off the harmonics to keep me legal when transmitting.

The comments about high speed ADC/DACs being expensive is dead on, they contribute highly to the cost of a nice SDR. The USRPs range in cost but can be as much as some HF basestations, though not as much as some of the crazy expensive contesting ones. Of course you also don't get power amplifiers and the knobs and buttons.

The BladeRF is another open hardware SDR worth taking a look at for some inspiration. At a block design level it's very similar to Ettus' B200 but uses a different RF frontend.

• https://nuand.com/bladerf.pdf

The Elecraft radios are heavily DSP based and their schematics are available here. The K3 and KX3 are nice designs and have all the traditional knobs. I'd love to drive one of these sometime. A fantasy of mine for a while has been to reverse engineer their protocol and make it a generic SDR control panel... one day

• http://www.elecraft.com/K2_Manual_Download_Page.htm#KX3

Cheers

[scatha]

Sounds like a fun job dkozel, Ettus makes some good gear and I really like the RFNoC architecture.

I've used the Ettus N210 with a BasicRX board in a few direct sampling receivers (mostly 138 MHz satcom), it's perfect because the analogue bandwidth of the ADCs (~500 MHz) is much greater than the ADC sample rate (100 MHz). This means that you can use sub-nyquist sampling to alias signals up to 500 MHz down to a more reasonable IF, no mixers required. You just need to supply the external filtering (BP cavity filter to combat the high-level VHF radio interferers) and LNA.

[ConKbot]

Here we go..  more bits, less bucks.. Just get a better sound card.. You may already have one.

QSD is inherently also a good band pass filter!  This is what Flexradio and Softrock and a lot of other quite economical quite good SDRs use. It really works well. Its like a rotary switch, click click click click.

http://www.google.com/search?q=quadrature+sampling+detector

http://www.google.com/search?q=tayloe+mixer

So your sound card is your ADC. And thats around the cheapest high quality ADC you are going to find anywhere and it works in the low frequency domain so the QSD converts your signals to that.  I and Q.

Agreed, it works well enough in SDRs that adc for the bw and dynamic range needed won't break the bank. But that's not what the ts asked for.

   The whole idea behind this is to filter and amplify your RF signal, and then directly sample the RF signal via the ADC, and everything leaving the DSP would exit through a DAC. All further processing (IF mixing, filtering, modulation, demodulation) is done digitally.

Rf in -> amp -> ADC.   With a rather impressive preselector that will go from 0-500MHz.

The usual trend of people being scared of RF and wanting to push all the scary bits into software.