The Future of SDR – Fat-Pipe vs. Thin-Pipe 19 - May - 2013Posted by k9zw in Amateur Radio, K9ZW, K9ZW Just Rambled.
Tags: K9ZW, K9ZW Just Rambled, PowerSDR, SDR, SDR Radio Server, SmartSDR, Thick-Pipe, Thin-Pipe
One thing that discussions, seminars, and banquets made very clear to me is the fundamental change in SDR design from Fat-Pipe design to Thin-Pipe design.
First what is meant by these terms?
Fat-Pipe is an SDR software and processing distribution with the on-board in-the-black-box hardware needing an external significant computer to make the radio work. Usually this is a PC running Windows or Linux/OS-X where the PC is doing many parts of the signal conversion. Analogue-Digital Conversions typically take place both in the radio box and in the PC.
Thin-Pipe is the SDR design where the in-the-black-box radio hardware does everything except the GUI (Graphical User Interface) and HMI (Human-Machine Interface). Thin-Pipe SDR radios often do not need a PC to operate “per say” but require the external device to change settings, display values and possibly handle audio I/O (some Thin-Pipe SDR designs can do their own direct audio I/O as well).
The darling of a experimenter building their own SDR from scratch or kit, Thick-Pipe SDR systems significantly are affected by the Host PC. The data transfer between the SDR and PC is multi path, complex and data intensive. Variations in the host PC capabilities, resource allocation/availability and every bit of software running are huge issues. Going remote is more easily handled by adding a Thin-Pipe between a remote PC and the host PC as the Thin-Pipe model better dealing with latency and thru-put complications.
It is non-trivial to open and sustain the broad multi-channel low-latency connection between the SDR with a remote Thick-Pipe PC.
Personal experience with PowerSDR Thick-Pipe configuration has shown the operating capabilities, present operating state of other software and the bluntly #%@$& issues of Windows driver, updates, conflicts, and foibles and endless set of issues.
So why has the move to Thin-Pipe first started now?
Processing Power and use of FPGA architecture.
When Thick-Pipe SDR designs rolled out the high end Host PC had 4 to 8 GFLOPS (billion floating point operations per second) capability against a typical all-in-the-box radio having perhaps 0.1-0.2 GFLOPS. The processing power in the SDR box was not significant. None of the hardware was hot enough to even worry about calculating GMACS (billion multiply-accumulate operations per second) which is arguably more important for SDR performance.
The new “SDR Radio Server” designs roll in onboard processors in the 100 plus GFLOPS range (right there with the biggest and meanest new PC processors) but with huge GMACS numbers running 300 plus.
The FPGA to a lay person like myself can basically be thought of like having 400-500 processor cores running parallel.
Basically this means that any PC is a weakling compared to the processing inside the box of a Thin-Pipe “SDR Radio Server” and the Host PC is an I/O manager if used at all.
In the Thin-Pipe design the Host PC puts the “pretty face” on the “SDR Radio Server” with displays of the SDR Radio Server’s settings, state and output. The Host PC also downloads to the SDR Radio Server the user’s audio (microphone/key/digital) , commands and simple functions like PTT (Push to Talk).
A Thick-Pipe SDR guru told me the data rate between SDR and Host PC differential between his state of the art imported Thick-Pipe SDR and the state of the art Thin-Pipe SDR design was 165 times heavier for the Thick-Pipe design even though it was working with less than 1/10th the sampled bandwidth of the Thin-Pipe SDR design.
This basically roughs up to the Thin-Pipe being content with the connectivity of a 2/3rds of a Skype connection, which is manageable (and affordable).
Will Thick-Pipe go away? Not likely. The cost point of a reasonable Thick-Pipe SDR complete station – especially if overall station performance envelope is not excessive – is attractive. Also many hobbyists are more comfortable tweaking their PC and PC’s software than are able to directly work with massive processing with FPGAs, which will keep the Thick-Pipe a favorite for the experimenter.
It is predictable that the leading edge of SDR performance will be Thin-Pipe – the brute force in the SDR Radio Server is so huge that the Thick-Pipe design doesn’t have a chance.
Now let’s categorize some real world SDR software by pipe type.
PowerSDR is Thick-Pipe in most implementations, with some implementations hybrid or Thin-Pipe control display I/O only. In most instances PowerSDR and interchangeable similar packages are Thick-Pipe. Connectivity is typically direct hardware hardwired, Firewire/1394, or high speed USB. Other mainly Thick-Pipe packages include Thick-Pipe include GSDR, CWExpert, SDRRadio and SDR#.
SmartSDR is an example of Thin-Pipe SDR implementation. Other quasi-SDR Thin-Pipe projects include WebSDR (which seems more a CAT receive only audio server).
Hope this has been helpful!