After what I'd read about the BG7TBL units and the fact that they seem to be based on an FLL rather than a PLL (accounting for the almost insignificant 0.15Hz frequency error in these units), I decided to take a fresh look at alternative "cost effective" GPSDOs and landed up on Leo's web site, scrutinising his GPSDO offerings which looked rather nice for the money.
If you're looking for cost effective GPSDO take a look at this
https://www.eevblog.com/forum/projects/my-u-blox-lea-6t-based-gpsdo-(very-scruffy-initial-breadboard-stage)/msg929949/#lastPost
I tested it on a breadboard and it works really well.
Hi Miti,
I was just reprising the whole of this thread (Again!) and had reached page 60 where you and the others had been discussing the various firmware update possibilities almost a whole year ago (My! How the time flies by!) when I noticed that you and two other guests were watching this thread so jumped to the last page (74!) to catch up with the latest postings.
I realised, after looking at the short thread you linked to, started by Gyro on his "My u-blox LEA-6T based fast locking GPSDO [experiment]", that I owed you a vote of thanks. The radioham rally is almost upon me (this coming Sunday the 28th at Blackpool) where I'm hoping to get some face time with actual electronics traders who sell this sort of stuff (OCXOs, ICs and pretty well anything that remotely relates to radio ham activities) and I've added a few more items mentioned by Gyro to my "Shopping List" which includes small 12 to 15 W triple rail +5, +/-12 or 15 volt psu boards (impossible to locate with google/ebay searches) amongst many other parts needed to upgrade my FY6600 and complete my GPSDO project.
Apologies! I've just realised that I've written yet another 'marathon sentence' that can't be read out aloud without risking asphyxiation.
I guess the art of packing "everything but the kitchen sink" into one "small sentence" doesn't translate quite so well from that of packing two adults and three small children and luggage into a 1965 Volkswagen Beetle for a 200 mile cross country one week's holiday trip that I still have fond memories of thirty years on.
Anyway, having refreshed my memory in regard of the perceived shortcomings of Feeltech's finest which we had all been obsessing over, I feel that the benefit of (my) hindsight might provide a useful perspective on this long drawn out saga of a thread.
Putting aside the several pages of commentary needed just to to explain in a nut shell the global geo-politics that's lead us to this point in time where we all seem to be headed for world catastrophe, it's quite clear that Chinese traders using Ebay haven't got the first clue about marketing to the western consumerist societies whose governments will, at the drop of a hat, prosecute wars in 'far away lands' in the pursuit of 'natural resources' to keep their source of income (tax paying consumers) sufficiently satisfied to avoid outright civil unrest and so maintain the status quo of the whole fatally flawed system upon which they rely for a trouble free life of luxury.
Sorry about mentioning "The Bleedin' Obvious" but I thought it best to remind ourselves of the underlying situation before slagging off our Chinese Suppliers of cheap and surprisingly good (in the main) test gear of which the FY6600 (and its successor, the FY6800) is a shining (fvsvo 'shining') example.
The most serious complaint we can legitimately raise, is the obvious lack of 'Good Faith" in the Feeltech camp when it comes to customer relations (eg. the lack of interest in putting right the issue of faulty firmware that was 'bricking' the FY6600 units cursed with version 3.0 firmware).
The rest of this product's shortcomings can be more sympathetically viewed as simply the inevitable consequence of cost cutting trade offs which should be taken almost for granted at the pricing levels involved. A crap product is better than no product at all at this sort of pricing level set to attract an otherwise ill served hobbyist market where such shortcomings can be worked around by a less demanding but more resourceful customer demographic. In this regard, Feeltech's targeting of their 'market' could hardly be bettered (putting aside the obvious shortcoming of customer relations in regard of the V3.0 firmware issue).
The main obsession appears to have been that of the half mains live 'touch voltage' that exists with any such kit powered from class II smpsus relying upon the mandated EMC bodge of the Y class capacitor to hold conducted common mode switching noise pollution at bay.
It turns out that the optimum solution to this problem (risk of ESD to a DUT) is simply to use a 3 wire mains cord with a 3 pole mains inlet socket to provide access to the PE to allow a low impedance (10KR) static drain connection to attenuate this half live touch voltage to half a volt or less whilst neatly avoiding the issue of earth loop induced interference at millivolt output levels (both millivolt DC and AC voltage offsets from induction and galvanic/thermocouple effects in the PE earth and mains supply wiring).
The alternatives of using analogue supplies or mains frequency isolating transformers to eliminate such 'touch voltage' effects without the use of a PE connection are ultimately doomed since it's impossible to completely eliminate capacitive coupling which is responsible for the problem in the first place. The risk may have been reduced (a 1000pF's worth from a Y capacitor reduced to a matter of 50 to 100pF's worth with a small 10 to 20 VA mains transformer - possibly much less with a special isolation transformer, the cost of which being a reduction of efficiency from 95 to 98 percent down to around 70%) but, nevertheless, a risk still remains.
Pursuing extremely low levels of capacitive coupling to mains voltage interference with special isolation transformers simply to avoid the use of a stiffer three core mains cable and a mains socket upgrade carries the penalty of extra heat dissipation within an already ill ventilated plastic case that's raising component temperatures perilously close to their upper limits for their life ratings (capacitors which may only be good for a few thousand hours at these temperatures versus several tens of thousands of hours by running just a mere 15 to 20 degrees cooler).
This issue of operating temperatures within the box takes on even greater importance when, for the sake of frequency stability, it is common practice to leave it switched on 24/7 so the avoidance of a low efficiency PSU becomes an ever higher priority requirement.
The existing smpsu board used by Feeltech turns out to be a much better optimised design than all of the similar three rail smpsu alternatives on offer via Ebay suppliers in that, unlike those offerings, the 12 volt rails have symmetric output capability (the Ebay offerings typically specify 1.5A on the positive rail whilst offering a mere 400mA or even less on the negative rail).
I've searched and searched Ebay till my eyes bled but haven't (so far) managed to find any dual output (let alone three rail) smpsu boards in the 10 to 20 watt range which offer symmetric current ratings on the 12 or 15 volt rails. In view of the utter simplicity in achieving such symmetry, as demonstrated in the Feeltech design, I'm rather at a loss as to why this should be the case.
The only shortcomings of the existing PSU board are the lack of screening and filtering and the insufficiency of voltage output on the "12 volt" rails[1]. I've so far managed to address the low 12 volt rail voltage issue by upgrading to proper high speed, low forward volt drop Shotky barrier rectifier diodes and adding a single turn winding to the transformer to buck the 5 volt winding to balance up the voltage distribution between the 5 and 12 volt rails (along with higher value smoothing caps - but only to a limit; too much capacitive loading causes the switching chip to go into an overload state).
The next step is to add 100nF ceramics across the input caps and mount the board into its own ventilated metal case with room to add additional LPF filtering on the output rails (using large value series input inductors with large value output shunt capacitors to suppress the HF switching ripple noise without inducing an overload response from the switching chip).
The ventilated metal case being a Faraday shield only requires an "earth connection" to the common ground rail and should avoid a low impedance connection to the PE - being a class II rated PSU, such a PE connection is not a safety requirement so the risk of ground loop induced interference can remain at bay, courtesy of the 10K static drain connection which nicely serves to suppress the half mains live touch voltage.
Bypassing this 10K drain resistor with a 100nF capacitor merely reintroduces the risk of earth loop induced HF interference and becomes somewhat of a folly in this case. Mention of which, when it comes to additional mains input filtering, we only need a series common mode inductor between the mains socket and the smpsu mains input live and neutral terminals. Live to neutral capacitors are ok but what we definitely don't require are any additional live and neutral to protective earth capacitive connections. Any such capacitors should have their grounding connection point left disconnected since they'll undo all the good work of the 10KR drain resistor in attenuating unwanted ground loop effects[2].
Ideally, we should likewise screen the main board if we wish to reduce direct radiation of the higher frequency signals used in testing HF radio gear but this becomes a task of far greater difficulty than that of merely containing the switching hash within the confines of a PSU screening box so is best left alone unless you're the type to see such effort as a challenge to be overcome at any cost.
The benefit to cost ratio in filtering and screening the PSU will imo, be a good two orders of magnitude greater than trying to do likewise for the main board so is worth the modest effort involved.
By all means, replace the existing PSU board with a ready made solution... if you can actually track one down, that is! (good luck with that search for unicorn droppings).
Turning to the issue of the original frequency accuracy and stability (lack thereof) of the smd XO chip, most of the problem seems to stem from the fact that it was located within just 10mm of the three LDO regulators which were, in my case, running at close to 70 deg C raising the XO chip to 50 deg C according to my IR thermometer. Now that I've reduced the 5v rail from the original boost to 5.49v, in my original effort to get the 12v rails above the 12v mark, back down to 5.07v, those LDO chips are probably now running a degree or three cooler.
It was no surprise to me when it was reported that just blowing over the XO chip caused a noticeable frequency shift. Indeed, it was this unconscionably high temperature which made me change my original plan to transplant the TCXO from the PCB it came supplied on (cheapest Ebay option for a 0.1ppm 50MHz TCXO) as a direct replacement to using the oscillator board as is and mount it at a jaunty angle over the 50mm 12v fan I had fitted into the base of the case and powered off the 5v rail in order to keep it in the 0.1ppm temperature zone and minimise power up warm up time.
Prior to that upgrade, trying to compare DIP XO frequency traces against the FY6600's frequency had been a game of "Chase Wil 'o the Wisp" due to that execrable smd XO chip's lack of anything that could be described as 'stability' by even the greatest stretch of the imagination.
Now, I am able to use it to compare the 8MHz output on the GPS module's PPS line, observing what I now understand to be the infamous "Saw Tooth" corrections being applied to the module's 48MHz TCXO generated clock pulses (20.8333ns jittering at anywhere from 10 per second to once every half minute or so).
Without such an upgrade to the FY6600, I'd have not been able to observe the actual mechanism by which a typical GPS module disciplines its internal clock to keep its 1PPS output synchronised to GPS time. IOW, this was a modification that was more than well worth my investment in both time and the 16 quid I'd spent on the "clock power board".
Another modification that's well worth the time and effort (unless you had no intention of generating MHz signals above the 5Vpp mark into high impedance loads[1]) is the replacement of the rather weedy THS3002i dual CFB opamp with a pair of THS3091/3095/3491 CFB opamp chips which will nicely eliminate the gross distortion when trying to produce sine wave outputs into 50 ohm loads at 20Vpp settings (10Vpp into 50 ohms) up to the 20MHz frequency limit before this limit is reduced to 5Vpp at frequencies beyond 20MHz (effectively disabling the use of the THS3002i or its replacements by bypassing it/them for signal levels at or below this 5Vpp limit).
The only remaining modification that comes to mind that isn't simply a matter of tweaking some preset trimmers on the main board, is that of reworking the 85 ohm attenuator pad into a 50 ohm pad which gets switched into the output by a relay whenever sub 500mV levels are selected. Feeltech are completely aware of this "Skoolboy Howler" since they've compensated for its effect in the High impedance case in the firmware. It cannot, of course, compensate both the high and 50 ohm terminated impedance states since it can only be one or the other. In this case, it's the Hi-Z case since this is unlikely to be noticed by anyone using the generator for audio frequency work since they rarely work with 50 ohm transmission lines and largely deal with feeding signals into 10K or higher impedances.
==========================================================================================
[EDIT 2020-04-16] With the benefit of 12 month's worth of hindsight, Feeltech's bean counter knew exactly what that 85 ohm attenuator pad nonsense was all about (reducing manufacturing costs at
any price).
It's obvious now that the design had been based on the industry standard practice of adding a 20dB attenuator and compensating by bumping the signal voltage up by one order of magnitude as you carry on selecting lower output levels in order to hold the effect of quantisation noise at bay for sub 50mV signal levels.
This was no 'Skoolboy Howler' with some clever firmware patch to compensate for a BoM error to get it to work in the Hi-Z loading case. It was simply an elimination of the more expensive E192 series resistors required to create a 20dB 50 ohm attenuator pad by replacing the 61.2 ohm shunt elements with 100 ohm cheapies and likewise, substituting the 249 ohm series resistor with another 510 ohm cheapie. The values had been selected solely to achieve precisely the same attenuation as the proper 20dB 50 ohm pad it had been designed to work with but only in the Hi-Z case.
The solution eventually proved to be as simple as "Just fit a 20dB 50 ohm pad, dummy!" after it was discovered that the FY6900 had exactly the same 85 ohm pad and testing the output of the primary opamp driving this attenuator proved that an order of magnitude step change was being applied at the 500/501mV threshold the attenuator was being switched into or out of the circuit with subsequent testing of my own FY6600 revealing exactly the same order of magnitude step change in voltage associated with the use of a 20dB attenuator just like every single signal/function generator on the market regardless of make.
The following text (which I've struck through) can therefore safely be ignored as pure speculation.
The lowest impedances they might deal with are likely to be 300 or 600 ohms where the expected discrepancy between 50 ohm source and 300 or 600 ohm sinks will be very close to what they'll observe in this case with an 85 ohm source. The truth of the matter will only come to light if the audio 'engineer' bothers to make the necessary and basic computations to verify the readings.
This annoying "Skoolboy Howler" with Feeltech's infamous 85 ohm attenuator (the FY6800 remains so afflicted despite the excellent opportunity that Feeltech wasted by not revising the BOM on this newer model's main board), is more likely to become swiftly apparent to those working at RF where the use of terminated transmission lines is more or less obligatory simply to avoid anomalous voltages where cable lengths are often a sizeable fraction of the wavelength of the frequencies involved.
Indeed, when working with impedance matched circuits, a standard technique for checking whether or not we have a proper impedance match is to observe the expected 6dB drop when connecting to a matched impedance (in this case, 50 ohms). The last thing any RF engineer expects to be doing is to account for an impedance change from 50 to 85 ohms just because the "Amplitude" setting had been reduced to 500mV or lower. Diagnosing problems is difficult enough when the test gear is within specification. Having it change its characteristic impedance just because of a change in output level is an unwanted and unnecessary evil that should never have been there in the first place.
I've had a go at fixing this annoyance but it turns out to be a lot more difficult to conjure up a 50 ohm impedance attenuator pad that matches the High impedance condition volt drop expected by the firmware bodge. I thought I'd gotten it properly sussed out but after tweaking the temporary trimpots I'd wired up to fine tune my calculated pass element value a lot more than I was expecting to, I'd landed up with a 45 ohm attenuator pad. Close but, as the saying goes, "No Cigar". It was, at the very least, an improvement but I'd like a much better improvement than that but, TBH, by the time I'd finished with all the calculations both pre and post the modification, I'd had more than my fill of attenuator calculations to last me for quite a while.
I'm still recovering from all that brain bursting effort even now, several weeks on. I figure that if I take a long enough breather, I might succeed with my next attempt at figuring out a new BOM solution. Basically, I'm just working to my main strength which is the art of procrastination. If anyone else fancies having a go at generating a new BOM for these attenuator pads, please, be my guest.
The key point to this is you have to aim for a 50 ohm attenuator pad which will match what the current pad gives under open circuit loading but only drop 6.02dB when driving a 50 ohm load instead of circa 8.6dB as it currently does. Somehow or other, I'd managed to screw up the calculations and I couldn't come up with consistent results each time I tried. It does look as though a 20dB 50 ohm attenuator pad had been the original design target but the firmware embedded correction to what I presume had been a BOM error means we can't fix it by simply dropping in a 20dB 50 ohm pad. This last conclusion where I'd assumed that a firmware kludge had been required to compensate for a "Skoolboy Howler" turned out to be tragically wrong. Dropping in a 20dB 50 ohm pad
turned out to be exactly the right solution as I could have determined if only I had thought to test my assumptions by verifying the voltage step change ratio with actual measurements.
[END_EDIT 2020-04-16]
==========================================================================================
To summarise, the modifications I think would be considered worth doing by most contributors in this thread (if they haven't done most of them already) are as follows:-
Upgrade the C8 mains connector to a C6 (allows a thinner cable to be used) or a C14 (tail wags dog effect) in order to connect the zero volt rail to the protective earth connection via a 10KR (or 1KR if 250 to 500mvac is still too much to stomach as an ESD risk) in order to kill off the half mains live voltage (50 to 90vac as measured with a typical DMM) on the BNC shields.
Don't directly connect the PE to the zero volt rail - the PSU is still a class II device
not requiring any such PE connection. We don't want to make the mistake that Feeltech did with the FY6800's earth connection by introducing an unnecessary grounding loop. This is a modification that everyone will benefit from regardless of their level of interest in frequencies beyond the audio frequency range.
Install a cooling fan - a small 50x50mm square by 10mm deep 12v fan run off the 5v rail will be more than sufficient to shift the heat out of a box that's not been vented with convective cooling in mind (tilting it on its prop stand actually aggravates the overheating effect!).
Replace the crappy commodity 50MHz smd XO chip with a half decent TCXO module, preferably, as others have done, by mounting it on a separate board remote from the original's hot spot on the main board. You could choose an OCXO if you're prepared to provide the additional 3 or 4 watts from an extra PSU board (if it's a 5v OCXO, you could utilise the innards of a cheap 2.1A USB wallwart for this task which will run even cooler once extracted from the confines of its unventilated wall plug shaped enclosure).
If you're planning on an opamp upgrade, then either modify the existing PSU board or else replace it with something better (good luck in finding a suitable replacement though). Apropos of which, the asymmetric nature typical of the current ratings on the +/-12 or 15 volt rails on most of the Ebay offerings could prove a boon if you're planning on fitting a 12v OCXO.
Upgrading the existing dual CFB opamp to a pair of the later spec THS 3091/3095/3491 opamps is a worthwhile modification unless your interests are essentially confined to low frequency audio work or you don't need "Amplitude" settings greater than 5Vpp.
That 85 ohm attenuator annoyance can, of course be worked around with an external 20dB attenuator but if you're regularly working with 50 ohm impedances, then replacing the resistors in that attenuator pad to convert it into a 50 ohm pad will provide a welcome benefit. Unfortunately, I can't offer much help there right now. My previous attempts only got me closer to the nominal 50 ohm than the original 85 ohm impedance match rather than close enough for me to tick it off my to do list. Just convert that 85 ohm attenuator into a 20dB 50 ohm pad by using E192 series resistors (or combinations of E24 or E48 series that will give you the required 148 and 61.2 ohm resistance values with some selecting on test if required). The solution
is that simple! I'd over-thought the problem but it turned out that the chief beancounter had told their designer to jump and the designer had simply asked "How high?"
I think that just about covers everything with regard to upgrading this sow's ear into a rayon purse. If I've missed anything, just speak your piece on the matter. I've not mentioned the 4ns jitter on square waves simply because it's inherent to the DDS technology used here and in other far more expensive AWGs, only alleviated in the later yet even more expensive famous brand named test kit. In view of its very low price, I think it's a relatively minor 'defect' we can all (learn to) live with.
JBG
[NOTES]
[1] Unless you're planning on upgrading the THS3002i dual CFB opamp with a pair of THS3091/3095/3491 opamp chips to eliminate distortion of the sine wave output at 20Vpp settings into 50 ohm loads in the frequency range 5 to 20 MHz, there's little point in doing the 12 volt rail voltage boosting mods. Indeed, if you don't select "amplitude" settings above the 5Vpp mark, the CFB opamp chip(s) never even get switched into circuit by the relays. This just leaves you with the business of upgrading the IEC C8 mains inlet socket to a 3 pole socket (C6 or C14 type) to provide a PE termination point for the grounded end of the 10 or 1 K ohm 'drain' resistor to eliminate the ESD risk posed by the Y capacitor.
[2] A perfect common mode choke would be made using bifilar winding but, aside from the mains voltage stresses on the inter-turn winding insulation, the separation of the live and neutral windings which eliminates this risk of insulation breakdown is also utilised to provide leakage inductance which is put to good use to create an effective transverse LPF by the use of additional capacitors across the live and neutral to suppress HF noise ripple voltages riding on top of the mains waveform using the live/neutral pair as a transmission line to reach vulnerable devices and unbalanced parts of the house wiring (eg. lighting switch drops) from where they can re-radiate as interference to wireless devices.
Most such ready made filters are designed for a LNE setup where a pair of Y caps are wired in series across the L&N X capacitor with the join intended to be connected to the PE, typically via a screening can connection. In this case, such Y capacitor earthing connections are redundant because they now become counterproductive to the need to keep mains half live touch leakage current to a minimum and avoid creating an unwanted earth loop at high frequencies.
If anyone wants to improve the common mode choke filtering between the PSU board and the mains socket, all that's needed is just a common mode choke or three, similar to the one already on the PSU board (scrapped smpsu wallwarts and the like are a good source). cascaded with X caps across the L & N connections of each section.
A shielded filtered mains socket where the Y caps are soldered to the PE connected shielding can still be used provided the earth tag is only used to connect to the 10KR drain resistor and not directly to the zero volt rail of the main board.
Having said all that, the real priority on 'filtering' lies with those LPFs between the DC voltage rails and the main board connections. If you're short on space within the PSU screening box, it's this LP filtering on the DC rails that gets priority. If you've room for both, you can leave the additional common mode choke filtering modification on the mains input side for later once you've verified the effectiveness of the DC filtering mod.