09 October 2022

PROJECT: Half-Life 2 AR2, Update #19 - Barrel Electronics

After many late nights and long train rides the AR2 Barrel electronics are complete! For those curious the electronics assembly consists of 4 PCBAs:


  • The Big Boss
  • Holds various switching & linear regulators that power servo & RED/WHT/RGB LEDs
  • LED brightness/colour & servo position are controlled with PCA9685PW, which itself is controlled via I2C
  • Lastly, as this is the main gateway to other PCBAs (one of which will be over a long cable...), the board has additional filtering & protection to improve EMI & ESD performance


    • Holds a pair of RGB LED chains for barrel glow effect
    • Interfaces to front & rear PCBAs
    • Has some really cool artwork on the silkscreen/overlay ;^)


    • Holds RGB & WHT LEDs for barrel glow & muzzle flash effect


    • Holds RGB & RED LEDs for barrel & shell glow effect

    And here's how the AR2 Barrel MECH & ELEC assembly looks like:

    Deep-dive into BARREL_MAIN-BOARD PCBA

    Just like with the RECEIVER_MAIN-BOARD I want to give a quick rundown of what I am happy with and what I know could be better (you know... have I had more time). But before getting into the nitty gritty here is a cool timelapse of the board being laid out in Altium:

    The Good

    1. GOOD, Solid 0V reference plane for high energy switching zones

    All of my switching regulators and complementary EMI filters are located on the bottom side, away from the "sensitive" digital zones on top layer. ~0.2mm below the bottom layer I have a solid 0V reference plane, ensuring that the high energy current traces have a closely coupled return path (think small current loop area, translating to lower emissions)

    And for those tracks that change their reference plane (as in jump from bottom to top layer), I make sure to use a 1N stitching capacitor to assist with the return path. More on this in the BAD section

    2. GOOD, Better suited bulk capacitors (package wise)

    Previously I was using 0603 sized ceramic capacitors for decoupling, filtering, & general bulk storage. A downside of such a "small" package (depending who you ask) is that with a typical X5R/X7R dielectric the capacitance will be dependant on bias voltage (and temperature), something most manufacturers don't show in their datasheet:

    NOTE: Electrolytic & tantalum capacitors also have this behaviour, and from memory polymer versions of the two are not as impacted. But as always you need to check the datasheet to know what to expect... as you can get drastically different performance with same dielectric material

    Is this a problem? As always, it depends on the application. But here are a couple of solutions/scenarios if I wanted to stick with using a ceramic capacitors:

    1. Use a dielectric that is "independent" on bias voltage or package temperature, like C0G/NP0. This is the way to go when precision is required (say an active filter), BUT be wary that getting a C0G capacitor that is >100N is going to get expensive
    2. Use a physically larger package (1206 instead of 0603) as having more bulk material assists with bias & temperature behaviour, BUT as you can expect this is at the cost of additional board space. Luckily for me I had plenty of that, so I just upped the package size:

    3. GOOD, Better suited EMI filter

    An improvement to my previous design the EMI filter topology has been changed from pi to T. Which is more appropriate when source & load impedances are LOW (like they are here) 

    The Bad

    1. BAD, Top side tracks have a poor reference plane

    I am very happy with my 0V reference plane (bottom side tracks), but I don't have the same enthusiasm for the reference plane used by the top side tracks... as the thing is incredibly choppy:

    What does this mean? Well I can expect to see increased emissions, as the return current for each trace can no longer run directly underneath. James Pawson of Unit 3 Compliance has a really good video on this, but below are some key slides to explain the issue:

    To help this discontinuity in return path I have sprinkled as many reference plane stitching capacitors as I can across the board. But now that I think of it I should have just poured 0V on the top side and stitched it to the internal 0V reference plane with a matrix of vias, spaced to reflect the highest frequency of concern

    2. BAD, LED connector positions could be better

    Though I am quite happy with the tight layout on the bottom side, once the LED related nets make their way to the top side the trace lengths become unpleasantly long due to the connector positions. Again I can expect increased emissions due to the larger loop area D:

    An easy solution would be to move the connectors (not possible), or throw more layers at the board

    3. BAD, EM zone boundary filters could still be better

    J101 could easily have a common mode choke on all 3 data lines, as each pin has a 0V conductor next to them...

    J102, P100, & P101 do not have a filter at all... So expect worse emissions (EM noise getting out) & susceptibility (EM noise getting in) performance here

    Schematic & PCBA

    And to close it all off, here are the BARREL_MAIN-BOARD schematics:

    And the PCBAs

    10 July 2022

    UPDATE: COVID, TV Cabinet, & Half-Life 2 AR2


    Life sure has a funny sense of humour, only last week I was having a chat with colleagues about how lucky I have been to have dodge the COVID bullet for so long... and well guess who caught the thing not long after D:

    Recovery wise things are progressing well. I got over the major flu-like symptoms relatively quickly (thanks to the vaccine and booster), and am now left with a blocked nose and a minor loss of smell

    TV Cabinet

    The waifu :3 and I have finalised our biggest (physically and workload wise) wood-working project to date, a TV cabinet. Just like with our Kitchen Shelves the plan is to fasten as many thing as possible with wood joints (vs say wood screws)

    A funny thing that did happened during the SOLIDWORKS assembly check is that we realised there may be a bit of an issue getting the thing in/out of the apartment...

    Lucky for us we realised that we can just stand the bottom cabinet upright and wedge it out of the lounge/entrance-hall bit by bit

    Half-Life 2 AR2

    So what does this all mean for the AR2? Well... we are aiming to finish the TV cabinet this year (and for a really good reason too), meaning work on the AR2 will be slow ; - ;

    BUT that does not mean I am giving up, and right now I am making some slow but steady progress on the AR2 Barrel assembly:

    25 June 2022

    PROJECT: Voice Training Buttons =^..^=

    Not long ago a mate of ours got a puppy and wanted to give "voice training" a go. For those unfamiliar, this is when you teach your pet to push a button corresponding to an activity they want to do, like in this amusing video:

    After a bit of online searching they saw that FluentPet had a rather expensive button kit, so asked us if we would be keen to give them a hand making a DIY "voice board". We said yes and decided to make a mini one for our cat with whatever spare parts remained :D

    So below is how our mini "voice board" compared to our CAD model, and for those curious how it was all wired up inside:

    16 April 2022

    PROJECT: Half-Life 2 AR2, Update #18 - RGB LED Driver Revision

    The last few months have been quite the rollercoaster for us, from Putin's unjust invasion of Ukraine, to me starting a new job... And only recently have I had the headspace to get back to the AR2

    So in this quick update I want to show yous the revised RGB LED driver configuration. Basically I will be halfling the barrel LED zones from 4 (2x3S & 2x4S) to 2 (2x7S). Doing so reduces the complexity of the barrel PCBA's (like halfling the RGB LED driver IC's from 4 to 2, and lowering the LED connector pin-count from 28P to 16P), which is a welcome change as each RGB LED driver (LT3496) goes for ~13AUD/pc in low quantities

    Now you might be asking, why not just drive all the 14 RGB LEDs in a big series configuration? Well I tried to simulate this scenario and soon realised that the forward voltage drop for the big LED chain would be too much for the LT3496 (who's SW pin is rated to 45V). I should also mention that I configure the OVP (Over-Voltage Protection) to 35V, so anytime the SW pin goes beyond this the IC shuts down for that switching cycle. Lastly below is the total forward voltage drop of each LED chain, and as you see in both cases we are over the 45V/35V limit:

    • RED ⇒ 2.6Vf * 14 ⇒ 36.4Vf-tot
    • GRN/BLU ⇒ 3.5Vf * 14 ⇒ 49.0Vf-tot

    While if we break up the chain into half (2x7S) we are nicely under the limit:

    • RED ⇒ 2.6Vf * 7 ⇒ 18.2Vf-tot
    • GRN/BLU ⇒ 3.5Vf * 7 ⇒ 24.5Vf-tot

    So with all that out of the way, here are the latest simulations of the 2x7S RGB LED configuration:

    And here is an updated power losses & expected temperature rise (per IC) table:

    IC Type RθJA, [°C/W] IC losses average, [mW] IC losses maximum, [mW] Temp above ambient, [°C]
    LT3496 Switching 34°C/W RED 68mW
    GRN 57mW
    BLU 57mW
    RED 113mW
    GRN 91mW
    BLU 91mW
    6°C to 10°C

    Couple of interesting things to note with above simulations/table:

    1. I highly suspect that the LTspice model I am using for the BLU/GRN LED is not the best as I am seeing massive ON/OFF & OFF/ON transients (like in orders of 100's of mA). I am certain that the peak current would be more closer to that of the RED LED simulation (~20mA above nominal), but to test this I plan to place an inline 0R jumper which I will then use to measure the actual peaks with a fancy oscilloscope at work
    2. The RED LED channel is showing a higher power loss, this is due to the RED LED current being double that of the GRN/BLU channel (40mA vs 20mA). So think greater I²R losses

    26 February 2022

    UPDATE: Ukraine 2022

    I am so utterly disappointed with Putin's decision to invade Ukraine. His lengthy dictatorship of Russia has put him out of touch with his people and the world as a whole, and it's so heartbreaking that he is willing to risk so much only to separate the many friends and families of two nations...

    The only sane voice in this conflict seems to be that of Ukraine's president, Volodymyr Zelenskyy, who released a well put plea for the Russian people to look beyond Putin's propaganda and see him for the tyrant he is

    28/02/22 Update

    A YouTube channel I follow called Супер Сус just released a video showing current life in Kyiv. In their usual videos they would show off abandoned metro tunnels or even explore the Chernobyl exclusion zone, all in a very colourful/expressionist way. But the thing that hit really hard in their latest video is seeing the hermetic doors being shut in the Kyiv Metro to protect the civilians on the other side... Like in their older videos whenever they come across a hermetic door they would explain that the door would close in the unlikely event of an attack on Kyiv... and here we are...

    05/03/22 Update

    For those that have friends & family living in Russia that are too deep in Putin's propaganda, below is a good website that shares a similar experience between a son living in Ukraine and a father living in Russia:



    The important thing is to keep the communication channel open. Don't be be angry and blame them for Russia's war/invasion of Ukraine. Make it clear that you are against Putin and his chokehold of the Russian people

    20 February 2022

    PROJECT: Half-Life 2 AR2, Update #17 - RGB LED Driver

    Guess who is back to crunching simulations in LTspice, this time trying to figure out how well the RGB LED driver works ;^)

    RGB LED, Configuration, & Driver Overview

    The RGB LED I have decided to use is the BROADCOM ASMT-YTD7-0AA02, which also comes with a diffused silicone cap to mix the emitted light

    The LEDs will be divided into 4 zones/strings inside the AR2 Barrel:

    Finally, after comparing LED drivers like HV9980, LT3597, LP5009... I decided to settle on the LT3496 to drive the above configuration. The thing that makes this driver special is that it has a Buck-Boost mode, which is a must when the nominal battery voltage hovers close to the total forward voltage of the LED sting (particular for the GRN & BLU channels). Below is the LTspice simulation for the RED string:

    A few key points about this LT3496 configuration:

    1. Reducing the current sense voltage (CTRL pin) from 100mV to 50mV lowers the peak LED current during ON-OFF transition from 110mA to 80mA (think reducing swing range of error amplifier). LEDs are rated for 100mA peak 100ns pulse and though the 110mA pulse was only for 25ns, I really wanted to be sure I am not hitting the 100mA maximum limit
    2. Reducing the switching frequency from 2.1MHz to 1.25MHz improves converter efficiency, as ON losses are lowered from 96mW to 76mW. Reducing the frequency beyond this brings diminishing returns, as losses are more or less ~70mW, while LED ripple current is increased (for same size inductor, 10μH)
    3. Chosen VC filter (22K & 470P) helps minimize LED current overshot and improves settling time (during OFF-ON transition)
    4. OVP resistor divider sets the LED string over-voltage protection to 35V. So if in the unlikely scenario that say a single LED fails in the RED string, the string will be safely shutdown to make sure it does not impact the GRN/BLU strings 

    RGB LED in Detail (Or Why We Need a Buck-Boost Mode)

    Remember when I said having a Buck-Boost mode is super useful when the nominal battery voltage hovers close to the total forward voltage of the LED sting? Well lets look at this in more detail... First off I will be using a 4S LiFePO₄ battery, so I expect the voltage to be:

    • 14.4V maximum
    • 13.6V nominal
    • 10.0V minimum

    I plan to drive each LED channel at 80% maximum DC current rating, so I expect the individual forward voltage to be:

    • RED 2.3Vf nominal @ 40mA, & a maximum of 3.0Vf
    • GRN 3.1Vf nominal @ 40mA, & a maximum of 3.6Vf
    • BLU 3.1Vf nominal @ 40mA, & a maximum of 3.6Vf

    So with a 3 series LED string the forward voltage will be:

    • RED 6.9Vf nominal & 9.0Vf maximum
    • GRN 9.3Vf nominal & 10.8Vf maximum
    • BLU 9.3Vf nominal & 10.8Vf maximum

    And with a 4 series LED string the forward voltage will be:

    • RED 9.2Vf nominal & 12.0Vf maximum
    • GRN 12.4Vf nominal & 14.4Vf maximum
    • BLU 12.4Vf nominal & 14.4Vf maximum

    Note how the LED string forward voltage spans the range of the battery voltage, meaning we can't just use a Buck or Boost regulator, as we will run into cases where battery voltage is too high/low for regulator to function. This is exactly where the Buck-Boost mode saves the day, as it can happily regulate the string voltage to required value :D

    Linear vs Switching (Or When Things Get Hot)

    So first of all we can't use a linear regulator to drive 4 series LEDs, unless we increase the battery voltage (as it would need to be ~2V higher than maximum forward voltage of the string). A linear regulator can just about drive 3 series LEDs, as in this scenario the battery voltage mostly gives enough headroom. With that said, lets compare how a linear regulator compares to a switching one

    NOTE: I am setting the LED string brightness with a 1kHz PWM waveform that is at 50% duty

    So using a switching regulator reduces the average dissipated power by ~70% (from 145mW to 40mW), and the ON dissipated power by ~80% (from 290mW to 56mW)... and that's just for the RED LED string/segment (as in not including the GRN/BLU string)! 

    Next lets expand the simulation to include the GRN/BLU LED driver losses and see what the expected IC (just the one, not the 4 I need to drive all barrel zones) temperature rise above ambient is:

    IC Type RθJA, [°C/W] IC losses average, [mW] IC losses maximum, [mW] Temp above ambient, [°C]
    LP5009 Linear 54°C/W RED 145mW
    GRN 120mW
    BLU 120mW
    RED 290mW
    GRN 239mW
    BLU 239mW
    21°C to 42°C
    LT3496 Switching 34°C/W RED 40mW
    GRN 46mW
    BLU 46mW
    RED 56mW
    GRN 69mW
    BLU 69mW
    5°C to 7°C

    And with all that waffle out of the way... we can see that using a switching regulator is the way to go, as:

    1. It is more energy efficient, resulting in a cooler AR2 Barrel ;^)
    2. It can actually drive 4 or more RGB LEDs in series, while being powered from same 4S LiFePO₄ battery

    09 January 2022

    PROJECT: Half-Life 2 AR2, Update #16 - Barrel Flex PCB

    So there are a couple of ways to add RGB LEDs to the AR2 barrel, and since I prefer to use SMD components I can either use a rigid PCB or a flex PCB. Normally I would go the rigid route, but this time I chose to use flex as I have never used them for hobby projects


    Next up I simulated the possible LED viewing angles in SOLIDWORKS (just like I did with the IR LEDs). From this I was able to see that the RGB LEDs need to have a viewing angle of at least 120° to have a good barrel coverage

    NOTE: Below simulation is super crude as it only looks at the 50% angular/spatial distribution (as in angle at which the brightness is at 50% maximum value), and does not account for the bright spot in the LED centre or the light diffusion within the material. Normally you would simulate all this with something like OpticStudio, but hey using SOLIDWORKS is better than nothing ;^)