|
I think that given all of the people experiencing blown components due to power supply issues (including myself), there is a case to be made for a redesign of the power supply. Something like an can replace 38 parts in the current design with 8, with significantly better performance in every way. The frequency of the part can be externally synchronized, but you can also simply configure it for 2.2 MHz, where every harmonic is a minimum of 400 kHz from any of the amateur bands covered by the QMX.
|
I think that given all of the people experiencing blown components due to power supply issues (including myself), there is a case to be made for a redesign of the power supply. Something like an can replace 38 parts in the current design with 8, with significantly better performance in every way. The frequency of the part can be externally synchronized, but you can also simply configure it for 2.2 MHz, where every harmonic is a minimum of 400 kHz from any of the amateur bands covered by the QMX.
Yes but...?
1. What about the eye-watering price? There are two versions in stock at Digikey, LTM8078IY and LTM8078EY. The price at 520-quantity?(1 tray) is $11.52 and $10.47 resp. Plus taxes of course... for comparison, since I have the two SMPS boards on the QRP Labs shop at $10? I'm sure you can imagine that the parts costs are way way less than one of these LTM8078; consider that the $10 price also includes the 6-layer PCB, the two female header connectors, SMD assembly factory costs, shipping, taxes, other costs etc.?
2. One of the main reasons for using a discrete component buck converter when the QDX Rev 3 PCB was designed, having the discrete component buck converter for the PIN diode forward bias generation, was that the global semiconductor crisis was in full swing, and adding an unusual part to the BOM seemed distinctly unattractive compared to a handful of discrete parts which are much easier to find. The QMX design inherited the buck converter from QDX. Whilst the global semiconductor crisis has receded somewhat we are still nowhere near back to the old days where you never ever thought about availability because you never had to. Looking at the available stock in Digikey (1050 of the EY and 3410 of the IY) isn't terribly encouraging in that regard.?
3. Being a BGA package, replacement would truly be beyond the reach of most of us here...
4. We are getting through building a larger and larger number of QMX now that the whole team are building them here. Originally, while we still had a long list of QDX, QCX-mini and QCX+ on the waiting list, only one (the most experienced and accurate) team member was building them, in order to build up experience of potential hazards. Of all the QMX I have yet seen, other than the Q103/Q104 Drain short (manufacturing problem) I have yet to see a failure that is not attributable to shorts, damaged components or other construction errors.?
73 Hans G0UPL
|
Hi Jonathan, Kees
These plots supplied by Jonathan look completely normal to me.?
The 3.3V rail experiences some disturbances during the boot-up process and this is completely normal; it is powered by the 78M33 linear regulator during that time via a 1N4148 diode. The diode voltage drop will naturally?show some variation depending on current draw, and current draw will be variable as the unit boots up various things.?
On Mon, Aug 28, 2023 at 6:27?PM Jonathan Burchmore < burch@...> wrote: Kees,
Here you go.? Measured on my working 9v QMX.
On:
Turn off with encoder press:
Hard turn off (in this case turning off the bench power supply, not actually pulling the plug):
Hope this helps,
Jonathan KN6LFB
|
It's a great part, other than the fact that it cost $12.80 for a tray of 1000, (About $20 per part in smaller quantities), is a single sourced part with no replacements if it becomes unavailable, and it's a ball grid array package, making it totally impossible for a purchaser of the QMX to ever replace it.??
-Steve K1RF
------ Original Message ------
Date 8/29/2023 12:08:18 AM
Subject Re: [QRPLabs] QMX - smoke - another C107/Q108 failure
I think that given all of the people experiencing blown components due to power supply issues (including myself), there is a case to be made for a redesign of the power supply. Something like an can replace 38 parts in the current design with 8, with significantly better performance in every way. The frequency of the part can be externally synchronized, but you can also simply configure it for 2.2 MHz, where every harmonic is a minimum of 400 kHz from any of the amateur bands covered by the QMX.
|
Hello Hans and Jonathan,
The plots require some questions to be answered in my mind.?
1) What is happening to 3.3V between t=80ms and t=200ms after Power ON ? Need more detail on the actual voltage seen.?Maybe Jonathan can expand that area on the scope so you see more 3.3V. detail ?
3) Both the 5V VCC and 3.3V VDD have a 470uF capacitor across the voltage outputs for voltage level stability, When the 78M33 Linear Regulator is being used it has ONLY a 2.2uF capacitor across the voltage output. Is that sufficient for voltage stability until the 3.3V SMPS is enabled ? .....or should it be larger to handle the initial power fluctuations during startup.
4) After 250ms and the 3.3V SMPS is turned ON, is the 78M33 Linear Regulator really turned OFF ??
Hans, I agree completely with not using the LTM8078 for the reasons you gave.
73 Kees K5BCQ
|
Jonathan,
After turning the power OFF and having the 3.3V VDD drop to 0V, how long does the 5V VCC stay at xxV ?
73 Kees K5BCQ?
|
The 1N4148 diode in series with the linear 3.3V output is a necessary evil to prevent backward current flow once the 3.3V SMPS rail comes up. However, it destabilizes the linear 3.3V rail as noted by the variation in voltage as the processor is executing its boot code. It is advisable to have a substantial capacitance on the processor’s power pins to minimize voltage. This should be at least 47uF - 100uF to ensure that the processor doesn’t experience an instantaneous dip below its minimum spec voltage which could cause the boot process to hang or go off into never-never land.
What happens on one unit is not representative of 1000’s of units so I hope the processor voltage during bootup is being tested on every unit prior to shipment.
Hi Jonathan, Kees
These plots supplied by Jonathan look completely normal to me.
The 3.3V rail experiences some disturbances during the boot-up process and this is completely normal; it is powered by the 78M33 linear regulator during that time via a 1N4148 diode. The diode voltage drop will naturally?show some variation depending on current draw, and current draw will be variable as the unit boots up various things.? On Mon, Aug 28, 2023 at 6:27?PM Jonathan Burchmore < burch@...> wrote: Kees,
Here you go.? Measured on my working 9v QMX.
On:
Turn off with encoder press:
Hard turn off (in this case turning off the bench power supply, not actually pulling the plug):
Hope this helps,
Jonathan KN6LFB
|
On Tue, Aug 29, 2023 at 08:29 AM, Kees T wrote:
1) What is happening to 3.3V between t=80ms and t=200ms after Power ON ? Need more detail on the actual voltage seen.?Maybe Jonathan can expand that area on the scope so you see more 3.3V. detail ?
Here's a tighter view: And zoomed in on the linear regulator portion. ?Vertical scale in both is 500mV/division with a -1.5V offset. Here's a similar trace with channel two showing LIN_REG_EN. ?Note the difference in scale and offset between the two channels. ?I didn't include VCC here because I don't have enough hands to hold the probes and not short anything out, sorry.
|
It looks like about 38 seconds or so, if I'm measuring this right.
Jonathan KN6LFB
|
On Tue, Aug 29, 2023 at 09:28 AM, Kees T wrote:
That LIN_REG_EN signal does not look right at all....it quits after about t=80ms.
It's possible I measured the wrong thing? ?I measured it from what I thought was pin 3 of the 3v3 SMPS board. ?The middle pin on the top. Jonathan KN6LFB
|
On Tue, Aug 29, 2023 at 09:45 AM, Kees T wrote:
That tells me that if you power ON again about 1-2 seconds after powering OFF, the 5V VCC rail is still at about 1.7V. Question is does the subsequent Power ON try to drive VCC above 5V ?
No, it did not appear to do so. ?If you look at my original scope traces you can see that VCC starts right around 1V because I had been turning the QMX on and off as I adjusted the scope trying to get a good view of things and it had not completely bled down. Jonathan KN6LFB
|
On Tue, Aug 29, 2023 at 09:58 AM, Kees T wrote:
You measured the right point. It just shows that the 78M33 3.3V VDD output is getting slapped around a little by several 2.2uF Tantalums on the board vs having a couple of 47uF Tantalums at the processor. Surprising that all those 2.2uF Tantalums on the board are apparently able to hold the 3.3V level for the full 250ms until the SMPS comes ON.
I don't think LIN_REG_EN is turning "off" at 80ms--I think what we're seeing is the handoff from the initial supply voltage (through PWR_ON?) to the 3.3v signal from the processor PD7 once it has booted up sufficiently. ?I seem to recall Hans discussing this in his FDIM presentation. Jonathan KN6LFB
|
On 29/08/2023 18:54, Jonathan Burchmore
wrote:
On Tue, Aug 29, 2023 at 09:58 AM, Kees T wrote:
You measured the right point. It just shows that the
78M33 3.3V VDD output is getting slapped around a little by
several 2.2uF Tantalums on the board vs having a couple of 47uF
Tantalums at the processor. Surprising that all those 2.2uF
Tantalums on the board are apparently able to hold the 3.3V
level for the full 250ms until the SMPS comes ON.
I don't think LIN_REG_EN is turning "off" at 80ms--I think what
we're seeing is the handoff from the initial supply voltage
(through PWR_ON?) to the 3.3v signal from the processor PD7 once
it has booted up sufficiently. ?I seem to recall Hans discussing
this in his FDIM presentation.
It seems to be in a half way state at that time, The linear
supply FET Q102 is on and so is Q110 which provides a load for the
SMPS. But Q111 is also on so the power can come from the linear or
the SMPS - or go into the SMPS and the 47R load.
Then when the SMPS pulls VDD above the 3V which the linear can
supply the SMPS takes over and LIN_REG_EN is pulled low.
Three state logic, on, half on and off.?
Chris, G5CTH
Jonathan KN6LFB
|
What is the mechanism that reduces the voltage on LIN_REG_IN during that half-way state from the supply voltage to 3.3V? ?It seems to me that the supply voltage passes through Q103 & Q105 (on the 5V board), then R101 on the 3v3 board and is directly connected from there to PD7 on the processor.
Jonathan KN6LFB
|
I think there's some good argument (personal enrichment not least
among them!) for using discrete parts. I like it, Hans!
The thing I found difficult was that both PS boards must
be present (and working) to bring up the microprocessor. (The 3.3
volt supply is dependent on +12 from PCB1...) It would have been
nice to be able to bring up the 3.3 volt board, then the 5 volt.
I can see that would have caused a problem with how parts
might've been distributed between the two boards. And if the 3.3
volts is working, really the 5 volt should be too. So it might be
phantom advantage!
That I see, the two major design challenges have been:
- Clearance between Q103/Q104, leading to solder bridges and
other challenges as people reworked the board. (I know that was
the root cause of my problems, anyway!)
- The impact of rapid changes (e.g., voltage spikes) in input
voltage as power comes up on V_IN. Maybe the best fix for this
would be to wait until the input supply stabilizes for a bit,
before engaging the switching supplies?
In software, we used to call a unique, elegant, piece of code a
"Neat Hack", a sort of backhanded compliment to the coder.
Hans, this unit is a Neat Hack in the realm of electronic (and
software) engineering. Lots of promise here-- You Go Guy! ;-)
Paul -- AI7JR
On 8/29/23 03:12, Hans Summers wrote:
toggle quoted message
Show quoted text
I think that given all of
the people experiencing blown components due to power supply
issues (including myself), there is a case to be made for a
redesign of the power supply. Something like an
can replace 38 parts in the current design with 8, with
significantly better performance in every way. The frequency
of the part can be externally synchronized, but you can also
simply configure it for 2.2 MHz, where every harmonic is a
minimum of 400 kHz from any of the amateur bands covered by
the QMX.
Yes but...?
1. What about the eye-watering price? There are two
versions in stock at Digikey, LTM8078IY and LTM8078EY. The
price at 520-quantity?(1 tray) is $11.52 and $10.47 resp.
Plus taxes of course... for comparison, since I have the two
SMPS boards on the QRP Labs shop at $10?
I'm sure you can imagine that the parts costs are way way
less than one of these LTM8078; consider that the $10 price
also includes the 6-layer PCB, the two female header
connectors, SMD assembly factory costs, shipping, taxes,
other costs etc.?
2. One of the main reasons for using a discrete component
buck converter when the QDX Rev 3 PCB was designed, having
the discrete component buck converter for the PIN diode
forward bias generation, was that the global semiconductor
crisis was in full swing, and adding an unusual part to the
BOM seemed distinctly unattractive compared to a handful of
discrete parts which are much easier to find. The QMX design
inherited the buck converter from QDX. Whilst the global
semiconductor crisis has receded somewhat we are still
nowhere near back to the old days where you never ever
thought about availability because you never had to. Looking
at the available stock in Digikey (1050 of the EY and 3410
of the IY) isn't terribly encouraging in that regard.?
3. Being a BGA package, replacement would truly be beyond
the reach of most of us here...
4. We are getting through building a larger and larger
number of QMX now that the whole team are building them
here. Originally, while we still had a long list of QDX,
QCX-mini and QCX+ on the waiting list, only one (the most
experienced and accurate) team member was building them, in
order to build up experience of potential hazards. Of all
the QMX I have yet seen, other than the Q103/Q104 Drain
short (manufacturing problem) I have yet to see a failure
that is not attributable to shorts, damaged components or
other construction errors.?
73 Hans G0UPL
|
Nothing.? The Linear continues to
produce 3.3V relative to it's 'GND' pin. if the supply produced by
the SMPS is more then it provides nothing.? All this is confused
by D102 andD103 which prevent the linear getting a voltage on its
output which it can't cope with.
Chris, G5CTH
On 29/08/2023 19:40, Jonathan Burchmore
wrote:
toggle quoted message
Show quoted text
What is the mechanism that reduces the voltage on LIN_REG_IN
during that half-way state from the supply voltage to 3.3V? ?It
seems to me that the supply voltage passes through Q103 & Q105
(on the 5V board), then R101 on the 3v3 board and is directly
connected from there to PD7 on the processor.
Jonathan KN6LFB
|
Until the microprocessor comes up, LIN_REG_EN will be pulled up
to +12 by R101. Once it's up and initialized, it will be fed by
the microprocessor output from it's 3.3 volt supply.
Does that fit what's observed?
Paul -- AI7JR
On 8/29/23 11:54, Chris wrote:
toggle quoted message
Show quoted text
Nothing.? The Linear continues to
produce 3.3V relative to it's 'GND' pin. if the supply produced
by the SMPS is more then it provides nothing.? All this is
confused by D102 andD103 which prevent the linear getting a
voltage on its output which it can't cope with.
Chris, G5CTH
On 29/08/2023 19:40, Jonathan
Burchmore wrote:
What is the mechanism that reduces the voltage on LIN_REG_IN
during that half-way state from the supply voltage to 3.3V? ?It
seems to me that the supply voltage passes through Q103 &
Q105 (on the 5V board), then R101 on the 3v3 board and is
directly connected from there to PD7 on the processor.
Jonathan KN6LFB
|
On Tue, Aug 29, 2023 at 11:54 AM, Chris wrote:
Nothing.? The Linear continues to produce 3.3V relative to it's 'GND' pin. if the supply produced by the SMPS is more then it provides nothing.? All this is confused by D102 andD103 which prevent the linear getting a voltage on its output which it can't cope with.
I did a poor job of stating my question. ?I agree with your assessment of the voltage regulator output. ?My question was about the LIN_REG_EN signal. ?If you look at the trace below, channel 2 is LIN_REG_EN as measured on the 3v3 board connector, i.e. downstream of R101 and Q101. ?It is initially around 7.5V, then drops to 3.3V, then to 0 when the SMPS comes online. ?I'm trying to understand the mechanism that pulls LIN_REG_EN FROM ~7.5V to 3.3V. ?Or for that matter why it is initially at 7.5V instead of the supply voltage of 9V. Secondly, since I was measuring the voltage right at the connector, it seems that the processor pin would also have been exposed to that 7.5V. ?Would that be a problem? ?Obviously the current would have been negligible due to R101. Jonathan KN6LFB
|
See my recent post here. Summary: Higher voltage is before the
microprocessor comes on line. The lower voltage is after it's
online and initialized, but not yet OK'd the switching power
supply.
On 8/29/23 12:24, Jonathan Burchmore
wrote:
toggle quoted message
Show quoted text
On Tue, Aug 29, 2023 at 11:54 AM, Chris wrote:
Nothing.? The Linear continues to produce 3.3V
relative to it's 'GND' pin. if the supply produced by the SMPS
is more then it provides nothing.? All this is confused by D102
andD103 which prevent the linear getting a voltage on its output
which it can't cope with.
I did a poor job of stating my question. ?I agree with your
assessment of the voltage regulator output. ?My question was about
the LIN_REG_EN signal. ?If you look at the trace below, channel 2
is LIN_REG_EN as measured on the 3v3 board connector, i.e.
downstream of R101 and Q101. ?It is initially around 7.5V, then
drops to 3.3V, then to 0 when the SMPS comes online. ?I'm trying
to understand the mechanism that pulls LIN_REG_EN FROM ~7.5V to
3.3V. ?Or for that matter why it is initially at 7.5V instead of
the supply voltage of 9V.
Secondly, since I was measuring the voltage right at the
connector, it seems that the processor pin would also have been
exposed to that 7.5V. ?Would that be a problem? ?Obviously the
current would have been negligible due to R101.
Jonathan KN6LFB
|
On 29/08/2023 20:20, Paul - AI7JR
wrote:
Until the microprocessor comes up, LIN_REG_EN will be pulled up
to +12 by R101. Once it's up and initialized, it will be fed by
the microprocessor output from it's 3.3 volt supply.
Does that fit what's observed?
Yes, but typically modern processors have a variety of control
methods that can be configured, one of which is an 'open
collector' style where it only pulls the line low and leaves it to
float high. Hans seems to have decided to pull it high.
Chris, G5CTH
Paul -- AI7JR
On 8/29/23 11:54, Chris wrote:
Nothing.? The Linear continues to
produce 3.3V relative to it's 'GND' pin. if the supply
produced by the SMPS is more then it provides nothing.? All
this is confused by D102 andD103 which prevent the linear
getting a voltage on its output which it can't cope with.
Chris, G5CTH
On 29/08/2023 19:40, Jonathan
Burchmore wrote:
What is the mechanism that reduces the voltage on LIN_REG_IN
during that half-way state from the supply voltage to 3.3V?
?It seems to me that the supply voltage passes through Q103
& Q105 (on the 5V board), then R101 on the 3v3 board and
is directly connected from there to PD7 on the processor.
Jonathan KN6LFB
|
Paul - AI7JR