Stanley are NOT what they once were....

Agree on the rest.
For more 'mechanical' tools, snap-on are nice but more money then god, probably not worth it unless you are a full time mechanic.

73 Dan.

I am not quite sure what exclusively for means in this context, and I will grant that it would not quite be turnkey, but if you took a hermes and pulled the anti alias filter, then preceded it with a suitable pair of BPFs and LNAs, you would come close by running it as a subsampler (The sample and hold bandwidth is ~700MHz).
The ADC noise figure does argue for quite a bit of front end gain, so you would need those filters.

Providing a VHF and UHF input path was one of the motivations for me trying to find the time to cook my own version.

73, Dan.

I am aware of the work, and was mainly bemoaning the missed opportunity to distribute a semi turn key radio with the required sampler, pad and relay preinstalled.

It is not an issue for you and me, but I get the feeling that the average purchaser of a 100D might be quite happy to upgrade the software and FPGA bitstream, but rather less happy to do the admittedly minor hardware hack required. 

I will probably end up getting a hermes, for all that what I would really like is an anglia logic board (The high speed IO port looks kind of interesting), as I am finding the time to spend with altium working up my own version is hard to come by.

Regards, Dan.

The annoying thing is that at almost no cost (AND without changing the PA in any way) the Anan radios could have provided a way to significantly improve the transmitter performance.

Adding a relay to switch the rx input to sample the forward power would have allowed either semi open loop adaptive predistortion (And could be done entirely in the baseband processing computer) or a cartesian loop (Really needs to be in the fpga), either of which should be good for 20dB improvement without breaking any sweat at all, and 30+db of improvement is possible.
-31dB ref PEP (From the Anan100D brochure) really is not great, -51 would be almost best in breed....
Even if nothing else, it would have allowed the waterfall to show the transmit spectrum as measured.

In the case of the 100D, it would have been possible to go one better, and sample both current and voltage at the aerial socket, which means that load mag Z and phase angle could be calculated. This opens up some interesting possibilities if one was to get smart with the drain power supply to the fets (This would obviously need PA changes).

Now granted, it is not a massive thing if starting with a hermes or similar to add that relay and appropriate attenuators and such, but it would have been nearly trivial to integrate onto the PA board in the ANAN series rigs and would have made all sorts of improvements only a software patch away.

73 Dan.

We are indeed having fun, particularly as I found a local guy who can mount fine pitch BGA and QFN for me on prototypes (hint, look for someone who repairs Playstation and X box units, the better end of these guys will have a proper BGA rework station).

My version is taking shape in Altium, a sort of Hermes variant along the lines of the 100D but set up for cartesian feedback on transmit and with the finals (pair of VRF151) drain supply derived from a buck converter so the drain supply can be made to track the envelope (and output impedance) for better power efficiency and less heat.

I have also moved the main anti alias filter into the RX BPF filtering so that subsampling will work for reception on the 2M & 70cm bands without needing a transverter (The 2208 has a sample and hold bandwidth out to 700Mhz). Transmit on these bands will need a different PA of course.

The neat thing about NIOS II is that while it is not itself a very good processor, it can tie into application specific FPGA logic, so expensive things like FFTs can be pushed out to dedicated hardware in the gate array. 

Now if only the bigger FPGAs (and especially the good AD chips) were a bit cheaper.....

Regards, Dan.

The flex appeared (at least in the hardware I saw) to have relay switched bandpass filters ahead of the ADCs, while the 100D uses a combination of the transmit LPF and switched HPFs on ADC1, with the second converter having only an anti aliasing filter....

The flex runs the ADCs at 250 Msamples/second vs 125M for the Hermes derived designs, possibly an issue on 6M where good aliasing performance is a big ask of the filters, this difference does however explain the larger slice count on the 100D as the FPGA area per slice will be about 1/4 - 1/2 that of that for the flex all else being equal.

The flex supports synchronous sampling at least for government users, but will apparently not support it for our market, this according to some discussion at the UK national rally last year (Something about ITAR regs in the US).
A pity, as electronically steered beam forming has interesting possibilities, especially if multiple receivers were widely geographically dispersed and locked to gps or similar (Think long baseline interferometry or HF radar).  The Anan hardware and software looks somewhat more amenable to having this sort of thing grafted on.

Finally, the Hermes/Anan rigs look to be hackable in a way the Flex is not.

73, Dan.

Software is hard, and worse is the next best thing to impossible to estimate reliably, real time software is an order of magnitude harder, realtime on a multitasking general purpose machine is harder again (And is just plain nasty to debug).

A big part of the issue is that the compute hardware out there is just so very variable, and even things like network cards (and their drivers) can be 'interesting', I have had a so called gigabit chipset start randomly dropping frames for no apparent reason, most unwelcome when trying to use UDP to move low latency audio around.

It might work here on my development box, but that is no guarantee that it will do the job on your packard bell laptop with a fan full of fluff that your mate has overclocked for you and has Norton AV hiding in the background fiddling with every system call I make, and storms of SMI exceptions trying to keep the poor thing from melting.

PCs provide a lot of cheap compute power, but low and deterministic latency is not their strong point.

Never underestimate the pain of going from a prototype to a product, especially when MS Windows is in any way involved.

What could be interesting is investigating the possibility of using a small dedicated compute board as a baseband processor for the anan/hermes radios, possibly itself with FPGA support or using CUDA or something, this would sit as a shim between the userland devices and create the the very high bandwidth I/Q data stream for the radios as well as doing the FFTs and suchlike, allowing the controls to run in a rather less time constrained environment.

I am thinking something like a TMS320C6xxx or SHARK, a Gb network port for the radios and a second network interface for the user network. The DSPs can handle the baseband processing without breaking a sweat, and can do it with negligible latency as they can work sample by sample in interrupt context rather then having to do blockwise processing like a PC does. As memory serves Hermes at least lets you select the amount of decimation in the CIC chain so it may well be practical to run at baseband bandwidths limited only by the network bandwidth to the radio.

This is essentially where the 6000 series is trying to go, but I want something hackable, including having the VHDL/Verilog be hackable, however good the supplied API may be, sooner or later you will hit something it will not let you do, and that sort of defeats the point of an SDR for me.

73, Dan.

Alas it is not quite that simple....

Frequency is the rate of change of phase, so to increase the frequency by say 5Khz you have to cause the RATE OF CHANGE of phase to increase by 2 * pi * 5000 radians per second.

Just adding a fixed phase shift does not do it, you have to make the phase shift increase with time.

Fortunately if you simply place an integrator between the modulation input and the phase modulator you get exactly the required behavior, so it is not an insurmountable problem, but that integrator does mean that the response is dependent upon the modulating frequency.

HTH.

73 Dan.

Part of this is actually down to changes in fashion among rig designers rather then anything to do with the move to dsp.

Basically, to work well an impulse noise blanker needs to see a really wide bandwidth (much wider then the desired audio).
Modern radios tend to put a roofing filter very early in the chain (often right after the first mixer), this filter does good things for overload performance, but converts fast rise time impulse noise into a slow rise time damped oscillation that is then very hard for anything to tell from a valid signal.
Fixing this really means driving the noise blanker from before the roofing filter, tricky in a rig using an upconverting first IF to provide continuous coverage. I guess that ideally you would wrap the noise blanker around the roofing filter with its level detector at the input side and the gating happening at the output, but with the woodpecker transmitter now glowing in the dark, it is not generally given the consideration it once was.

DSP filtering (Especially when turned up to stun) can sound really nasty, but if you have the controls turned that far up is is probably because you are trying to dig something out of the noise, and you probably could not even hear it without the processing.
 
Now I will grant that there are DSP rigs out there where the AGC design falls to bits in the presence of multiple strong signals, which really comes down to poor design, but there have always been poor quality radios.

73 Dan.

Ah, at frequency phasing method, doable in DUC/DDC but always going to be a little interesting in an analogue mixer based SDR, better to use a fixed IF if doing that.

73 Dan.

SDR is not magic, if you want good spurious performance you still have to ensure linearity, including linearity in the digital domain, just like you do for a conventional rig.

To give one simple example, multiplying two N bit numbers (Say implementing a mixer) gives a result having a length of 2N, if we then want to shorten that to say just the upper N bits, there are two ways we can do it.

1: Just take the top N bits to the next stage - Unfortunately this results in non linearity as anything smaller then 1 LSB of the output will be truncated to zero, additionally weak components that happen to force the quantization between levels will be effectively mixed with other signals present and will cause discreet mixer spurs.

2: Add a pseudo random noise having triangular probability to the 2N long word before truncating, this adds a constant noise floor but makes the truncation perfectly linear, an enhancement is to shape this noise in the frequency domain to reduce the energy in whatever our eventual passband will be, but in either case it takes much FPGA area to pull off especially in the high rate bits of a DUC/DDC design.

It sounds however like the problem flex had was probably something mundane and hardware related rather then being a problem in the vhdl/verilog.

Regards, Dan.

Add to those the various word length issues where folk have truncated without appropriate dithering (Tends to be more of an issue in DDC where correctly handling word length reduction down the CIC decimation chain can be a headache).

In general any word length reduction should be accompanied by appropriate dither if you wish to avoid weird spurs in places you would not necessarily expect, but doing this in the high rate stages is **expensive** in terms of FPGA area.

This is discussed in the literature relating to the use of fixed point math for DSP, sometimes it appears in the chapter dealing with word length in IIR filters. 

73 Dan.

Don't forget that heating in the wiring is proportional to average of (current squared) so a short term 20A at say 10% duty cycle produces much more heat then 2A continuous would.

If we assumed say 0.1 ohms loop resistance then during that 20A pulse we are dropping 2V in the wiring and dissipating 40W in the cables (For 4W average dissipation at 10% duty), the steady state 2A is dropping 0.2V and dissipating 0.4W for the same average current in the load. 

Fitting significant local energy storage may help, but how much strongly depends on how much voltage drop you have under load, and what the power back off ratio for the modulation mode in use is in terms of the current envelope (quite different to the RF power envelope with most simple minded PA designs).

In any case, lighter sockets are notoriously poor, better to use almost anything else (Speakon loudspeaker connectors are nice).

Regards, Dan.

The thing about an RF amp is that by and large we do NOT care about the shape of the RF (That is what lowpass filters are for!), but we care deeply about the shape of the envelope.

There are good ways to make an efficient 'digital' amp for HF, but they typically do not do direct PWM and usually require additional signals from the exciter, further they are often rather narrowband (See the class E stuff beloved of the AM Phone crowd). HF extends across ~4 octaves, which is way too wide for easy direct PWM or such (Harris have a PWM rig for shortwave broadcast but it is not exactly frequency agile).

One efficiency trick that occurred to me is to note that most amps are specified to deliver full power into 1.5:1, meaning that the drain match has to be such that full power will be delivered into 75 ohms, now that means that for the same power into 50 ohms the transistors are dissipating quite a lot of heat because the AC component of the drain voltage is smaller then optimal. However if one placed an impedance bridge at the output of the PA then feed |V|/|I| back to a switchmode power supply, the drain voltage could be set to only that required for the load impedance actually being presented (Especially helpful when running a rig that can do full power into 1.5:1 hi Z at 1.5:1 low Z....).
Result: less heat and efficiency much closer to that typical of a tube amp with its variable match (at least for resistive loads).
I have been getting some promising results from feeding the envelope to a buck converter so as to have the supply partially track the envelope, but that really needs advanced feedback techniques to compensate the resulting phase modulation, at which point you have a transmitter as opposed to an amplifier. 

73 M0HCN.

Actually, damp sponge is not a good way to clean a tip, better to use brass wool as it does not impose the thermal shock that can strip the plating from the bit, and yea tip cleaning on the jeans is a BAD plan.

In other notes, it works better when it's plugged in!

... Take one large wirewrap board full of 4000 series cmos, some of it sequential, place said card into rack in industrial control system, start getting phone calls at odd hours about automation failures.

After much swearing, and far too many late nights, eventually find that the power supply switch was set to O-F-F and the thing was being powered from the logic lines via the input protection diodes! Worked fine until one too many inputs went low where upon the thing would loose its internal state.
Move switch from O-F-F to O-N and everything worked perfectly.

Regards, Dan.