content/hp34970a_5.rst

==========================================
 HP 34970A Data Acquisition Unit - part 5
==========================================

:Author: David Douard
:Category: Electronics
:Tags: HP, 34970A, HP34970A, DMM, repair, test equipment
:series: HP 34970A repair
:series_index: 5


After a loooong pause on this project, I have recently been able to work on it
again. In `the previous post in this series <{filename}/hp34970a_4.rst>`_, I've
started to implement a replacement board prototype based on a an NUCLEO board
and an SSD1322-based OLED display.

Meanwhile, I have also almost completely described the `communication protocol
<{filename}/hp34970a_protocol.rst>`_ between the main unit and the front panel.

In this episode, I'll describe the next steps: make the circuit diagram and the
PCB for the replacement board.


Circuit diagram
===============

The circuit diagram is rather simple:

- an STM32 MCU (I initially chose the `STM32F303RE
  <https://www.st.com/en/microcontrollers-microprocessors/stm32f303re.html>`_,
  but finally used an STM32F303RD just because the former was out of stock when
  I made my command at mouser.) I wanted an STM32 MCU with USB support and
  reasonnable amount of flash and ram just not to have to worry about these
  while writing the firmware.

- a 8MHz crystal so the clock config used on NUCLEO boards can be used as is.

- a step-down DC-DC converter to go from the +18V coming from the main unit
  to +5V to power the OLED display and the some chips on the board; I've chosen
  a `TPS560430 <https://www.ti.com/product/TPS560430>`_ from TI,

- a 5V->3.3V LDO regulator to power the MCU, an `MCP1811AT-033
  <https://www.microchip.com/wwwproducts/en/MCP1811>`_ by Microship,

- a level shifter for the communication lines bertween the MCU and the main
  unit (5V vs. 3.3V logic). For this, instead of going through individual
  Mosfet transistors, I've chosen the `LSF0204PWR
  <https://www.ti.com/product/LSF0204>`_ from TI, to simplify a bit the
  assembly (this chip really is not much more than a bunch of transistors in a
  TSOP package).

Here is the diagram I ended up with:

.. image:: {static}/images/hp34970a/replacement-front-panel-schematic.svg
   :alt: Schematic for a replacement front panel for the HP 34970A
   :class: image-process-large-photo

PCB design
==========

I've used librecad_ to draw the edge profile of the PCB, after having carefully
measured the broken PCB, and a kicad plugin to generate nice keypad footprints.
(I had to adjust the generated footprint by hand to make it more usable.)

The end result looks like:

.. image:: {static}/images/hp34970a/front_panel_replacement_pcb_3d.png
   :alt: 3D view of the PCB for a replacement front panel for the HP 34970A
   :class: image-process-large-photo

The 2 buttons on the left are the Reset and the DFU boot mode. The header on
the right is the SWD port.

After a few days, the PCB arrived (from the usual places):

.. image:: {static}/images/hp34970a/bare_pcb_front.jpg
   :alt: the PCB for a replacement front panel for the HP 34970A
   :class: image-process-large-photo

And it fits just fine in the front panel enclosure:

.. image:: {static}/images/hp34970a/pcb_test_fit.jpg
   :alt: test fit of the PCB for a replacement front panel for the HP 34970A
   :class: image-process-large-photo


Unfortunately, I made a few mistakes in this design (see below) bit I also made
the big mistake of swaping the marking of R11 and R12, which defined the
feedback divider setting the output voltage of the DC-DC converter. As a
result, I fried a LDO and a STM32F303RD chips!

But once replaced correctly, is began to work:


.. image:: {static}/images/hp34970a/pcb_dfu_connection.jpg
   :alt: test USB connection in DFU mode
   :class: image-process-large-photo

After installing the OLED display and adjusting the previous verison of the
code (which was designed for the NUCLEO F446RE board), I could compile it for
my custom board and make it run:

.. image:: {static}/images/hp34970a/pcb_first_test.jpg
   :alt: test of the display circuit
   :class: image-process-large-photo

Firmware design
===============

At this point, I tried to make the CDC USB driver work on the board, but I
failed to do so. I realized that the USB stack is `actually missing
<https://forums.mbed.com/t/usbdevice-support-for-stm32f3/10685>`_ for this MCU
in MBed OS 6. (update: there is a `PR
<https://github.com/ARMmbed/mbed-os/pull/13874>`_ waiting to be accepted and
merged to improve support for the STM32F303 family).

Unfortunately, I could not make the debug pin of the SWD port work either. I
can upload a new firmware juste fine via this SWD port (in addition to the DFU
mode via USB), but I have been ablt tuse the SWO ping to get a console to see
print statements.

In last resort, I ended up soledring thin enameled wired directly on the MCU's
pin to get access to of the available UARTs as console/debugger.

At least this worked just fine, and I could then work on the code much more
easily.

I also had to use my DSLogic logic analyzer to debug a few things and get to a
point that make the front panel usable:

.. peertube:: cc6f4327-703f-42f6-8157-03cd0a987387

In this version of the code are many remaining problems, the biggest one being
the fact the display flickers a lot. In the code of the formware, I have a
"frame"-buffer in which the main loop draw the information to be displayed, and
a thread that copy this buffer to the actual display once every 30ms. And in
the first version display above, there was no DMA involved in tranferring the
buffer into the display unit.

Activating the DMA did help a bit, but was not enough to get rid of the
glitches. The real problem was there was no lock around that buffer, so the
display refresher thread was making the transfer at any time, including in the
middle of the drawing proceure. Adding a lock around this buffer finally fixed
the problem.

Another problem I had in the first version of the firmware was the fact a
keypress event could be sent to the main unit during a transmission of a packet
by this later, which mess the transmission protocol quite a bit. This problem
is now (mostly, not 100% sure there is no edge case left) fixed.

Here is an example of this situation occuring (screenshot of a `PulseView
<https://sigrok.org/wiki/PulseView>`_ session):

.. image:: {static}/images/hp34970a/packet-collision.png
   :alt: transmission collision
   :class: image-process-large-photo

In the example above, one can see the front panel is starting a transmission
(sends a 0x66 on the RX line), but while this byte is being sent, the 0x66
start of transmission byte is also sent by the main unit.

Not sure yet how to fix this, either by making sure that this situation does
not occur (which is most unlikely not feasible), or by impriving the transmission
protocol handling state machine in my firmware to be aware that this may happen
and be resilient against it.

But in the end, I've now reach a point where the firmware begins to be really
usable:

.. peertube:: 172bf449-eba7-4429-8840-ab9b2d5398f8


Design errors
=============

Making this circuit I made a few mistakes and errors.

Wrong MCU
---------

I've not paid enough attention when I chose the MCU. The STM32F303RE I chose is
more then fine on a specifications point of view, however, since I am writing
the formware using MBedOS, I did pay attention that the USB FS stack is not yet
supported for this MCU on MBed OS 6.

My initial idea was to provide a serial communication protocol with the front
panel to be able to upgrade the firmware easily or get the current display
content, but this is for now not possible. Will have to stick to GPIB for now.


No proper USB VBUS handling
---------------------------

I decided not to connect the VBUS power line from the USB socket, but it would
have made my life a bit easier to connect it via the usual protection, so the
testing / uploading the firmware could have been a bit easier until the front
panel wasa plugged back to the main unit.

Even so, it would be nice to be able to upgrade the firmware without having to
plug the whole unit.

The real problem of not having connected the VBUS signal is USB detection. If I
want to be able to implement a USB connection, I need to be able to detect
wether the USB connector is plugged, and the simplest way of doing that is to
detect the presence of the VBUS signal.

No diagnostic/status LED
------------------------

I should have added a few diagnostic/status LEDs: there are plenty of GPIO left
unused on the MCU, having those visble indocators is alway handy.

No extra UART
-------------

I should have routed at least one unused UART from the MCU to a pair of
headers: it way easier not to populate such a header than to hack small wires
directly on the MCU tget a decent console port...


.. _librecad: https://github.com/LibreCAD/LibreCAD