southpark cartoon

Some thoughts on Storytelling as an engineering teaching tool

Every week at work on Wednesday afternoons we have the SRE ops review, a relaxed two hour affair where SREs (& friends of, not all of whom are engineers) share interesting tidbits that have happened over the last week or so, this might be a great success, an outage, a weird case, or even a thorny unsolved problem. Usually these relate to a service the speaker is oncall for, or perhaps a dependency or customer service, but we also discuss major incidents both internal & external. Sometimes a recent issue will remind one of the old-guard (of which I am very much now a part) of a grand old story and we share those too.

Often the discussion continues well into the evening as we decant to one of the local pubs for dinner & beer, sometimes chatting away until closing time (probably quite regularly actually, but I'm normally long gone).

It was at one of these nights at the pub two months ago (sorry!), that we ended up chatting about storytelling as a teaching tool, and a colleague asked an excellent question, that at the time I didn't have a ready answer for, but I've been slowly pondering, and decided to focus on over an upcoming trip.

As I start to write the first draft of this post I've just settled in for cruise on my first international trip in over six months[1], popping over to Singapore for the Melbourne Cup weekend, and whilst I'd intended this to be a holiday, I'm so terrible at actually having a holiday[2] that I've ended up booking two sessions of storytelling time, where I present the history of Google's production networks (for those of you reading this who are current of former engineering Googlers, similar to Traffic 101). It's with this perspective of planning, and having run those sessions that I'm going to try and answer the question that I was asked.

Or at least, I'm going to split up the question I was asked and answer each part.

"What makes storytelling good"

On its own this is hard to answer, there are aspects that can help, such as good presentation skills (ideally keeping to spoken word, but simple graphs, diagrams & possibly photos can help), but a good story can be told in a dry technical monotone and still be a good story. That said, as with the rest of these items charisma helps.

"What makes storytelling interesting"

In short, a hook or connection to the audience, for a lot of my infrastructure related outage stories I have enough context with the audience to be able to tie the impact back in a way that resonates with a person. For larger disparate groups shared languages & context help ensure that I'm not just explaining to one person.

In these recent sessions one was with a group of people who work in our Singapore data centre, in that session I focused primarily on the history & evolution of our data centre fabrics, giving them context to understand why some of the (at face level) stranger design decisions have been made that way.

The second session was primarily people involved in the deployment side of our backbone networks, and so I focused more on the backbones, again linking with knowledge the group already had.

"What makes storytelling entertaining"

Entertaining storytelling is a matter of style, skills and charisma, and while many people can prepare (possibly with help) an entertaining talk, the ability to tell an entertaining story off the cuff is more of a skill, luckily for me, one I seem to do ok with. Two things that can work well are dropping in surprises, and where relevant some level of self-deprecation, however both need to be done very carefully.

Surprises can work very well when telling a story chronologically "I assumed X because Y, <five minutes of waffling>, so it turned out I hadn't proved Y like I thought, so it wasn't X, it was Z", they can help the audience to understand why a problem wasn't solved so easily, and explaining "traps for young players" as Dave Jones (of the EEVblog) likes to say can themselves be really helpful learning elements. Dropping surprises that weren't surprises to the story's protagonist generally only works if it's as a punchline of a joke, and even then it often doesn't.

Self-deprecation is an element that I've often used in the past, however more recently I've called others out on using it, and have been trying to reduce it myself, depending on the audience you might appear as a bumbling success or stupid, when the reality may be that nobody understood the situation properly, even if someone should have. In the ops review style of storytelling, it can also lead to a less experienced audience feeling much less confident in general than they should, which itself can harm productivity and careers.

If the audience already had relevant experience (presenting a classic SRE issue to other SREs for example, a network issue to network engineers, etc.) then audience interaction can work very well for engagement. "So the latency graph for database queries was going up and to the right, what would you look at?" This is also similar to one of the ways to run a "wheel of misfortune" outage simulation.

"What makes storytelling useful & informative at the same time"

In the same way as interest, to make storytelling useful & informative for the audience involves consideration for the audience, as a presenter if you know the audience, at least in broad strokes this helps. As I mentioned above, when I presented my talk to a group of datacenter-focused people I focused on the DC elements, connecting history to the current incarnations; when I presented to a group of more general networking folk a few days later, I focused more on the backbones and other elements they'd encountered.

Don't assume that a story will stick wholesale, just leaving a few keywords, or even just a vague memory with a few key words they can go digging for can make all the difference in the world. Repetition works too, sharing many interesting stories that share the same moral (for an example, one of the ops review classics is demonstrations about how lack of exponential backoff can make recovery from outages hard), hearing this over dozens of different stories over weeks (or months, or years...) it eventually seeps in as something to not even question having been demonstrated as such an obvious foundation of good systems.

When I'm speaking to an internal audience I'm happy if they simply remember that I (or my team) exist and might be worth reaching out to in future if they have questions.

Lastly, storytelling is a skill you need to practice, whether a keynote presentation in front of a few thousand people, or just telling tall takes to some mates at the pub practice helps, and eventually many of the elements I've mentioned above become almost automatic. As can probably be seen from this post I could do with some more practice on the written side.

1: As I write these words I'm aboard a Qantas A380 (QF1) flying towards Singapore, the book I'm currently reading, of all things about mechanical precision ("Exactly: How Precision Engineers Created the Modern World" or as it has been retitled for paperback "The Perfectionists"), has a chapter themed around QF32, the Qantas A380 that notoriously had to return to Singapore after an uncontained engine failure. Both the ATSB report on the incident and the captain Richard de Crespigny's book QF32 are worth reading. I remember I burned though QF32 one (very early) morning when I was stuck in GlobalSwitch Sydney waiting for approval to repatch a fibre, one of the few times I've actually dealt with the physical side of Google's production networks, and to date the only time the fact I live just a block from that facility has been used at all sensibly.

2: To date, I don't think I've ever actually had a holiday that wasn't organised by family, or attached to some conference, event or work travel I'm attending. This trip is probably the closest I've ever managed (roughly equal to my burnout trip to Hawaii in 2014), and even then I've ruined it by turning two of the three weekdays into work. I'm much better at taking breaks that simply involve not leaving home or popping back to stay with family in Melbourne.
southpark cartoon

Building new pods for the Spectracom 8140 using modern components

I've mentioned a bunch of times on the time-nuts list that I'm quite fond of the Spectracom 8140 system for frequency distribution. For those not familiar with it, it's simply running a 10MHz signal against a 12v DC power feed so that line-powered pods can tap off the reference frequency and use it as an input to either a buffer (10MHz output pods), decimation logic (1MHz, 100kHz etc.), or a full synthesizer (Versa-pods).

It was only in October last year that I got a house frequency standard going using an old Efratom FRK-LN which now provides the reference; I'd use a GPSDO, but I live in a ground floor apartment without a usable sky view, this of course makes it hard to test some of the GPS projects I'm doing. Despite living in a tiny apartment I have test equipment in two main places, so the 8140 is a great solution to allow me to lock all of them to the house standard.


(The rubidium is in the chunky aluminium chassis underneath the 8140)

Another benefit of the 8140 is that many modern pieces of equipment (such as my [HP/Agilent/]Keysight oscilloscope) have a single connector for reference frequency in/out, and should the external frequency ever go away it will switch back to its internal reference, but also send that back out the connector, which could lead to other devices sharing the same signal switching to it. The easy way to avoid that is to use a dedicated port from a distribution amplifier for each device like this, which works well enough until you have this situation in multiple locations.

As previously mentioned the 8140 system uses pods to add outputs, while these pods are still available quite cheaply used on eBay (as of this writing, for as low as US$8, but ~US$25/pod has been common for a while), recently the cost of shipping to Australia has gone up to the point I started to plan making my own.

By making my own pods I also get to add features that the original pods didn't have[1], I started with a quad-output pod with optional internal line termination. This allows me to have feeds for multiple devices with the annoying behaviour I mentioned earlier. The enclosure is a Pomona model 4656, with the board designed to slot in, and offer pads for the BNC pins to solder to for easy assembly.



This pod uses a Linear Technologies (now Analog Devices) LTC6957 buffer for the input stage replacing a discrete transistor & logic gate combined input stage in the original devices. The most notable change is that this stage works reliably down to -30dBm input (possibly further, couldn't test beyond that), whereas the original pods stop working right around -20dBm.

As it turns out, although it can handle lower input signal levels, in other ways including power usage it seems very similar. One notable downside is the chip tops out at 4v absolute maximum input, so a separate regulator is used just to feed this chip. The main regulator has also been changed from a 7805 to an LD1117 variant.

On this version the output stage is the same TI 74S140 dual 4-input NAND gate as was used on the original pods, just in SOIC form factor.

As with the next board there is one error on the board, the wire loop that forms the ground connection was intended to fit a U-type pin header, however the footprint I used on the boards was just too tight to allow the pins through, so I've used some thin bus wire instead.



The second major variant I designed was a combo version, allowing sine & square outputs by just switching a jumper, or isolated[2] or line-regenerator (8040TA from Spectracom) versions with a simple sub-board containing just an inductor (TA) or 1:1 transformer (isolated).



This is the second revision of that board, where the 74S140 has been replaced by a modern TI 74LVC1G17 buffer. This version of the pod, set for sine output, uses almost exactly 30mA of current (since both the old & new pods use linear supplies that's the most sensible unit), whereas the original pods are right around 33mA. The empty pods at the bottom-left are simply placeholders for 2 100 ohm resistors to add 50 ohm line termination if desired.

The board fits into the Pomona 2390 "Size A" enclosures, or for the isolated version the Pomona 3239 "Size B". This is the reason the BNC connectors have to be extended to reach the board, on the isolated boxes the BNC pins reach much deeper into the enclosure.

If the jumpers were removed, plus the smaller buffer it should be easy to fit a pod into the Pomona "Miniature" boxes too.



I was also due to create some new personal businesscards, so I arranged the circuit down to a single layer (the only jumper is the requirement to connect both ground pins on the connectors) and merged it with some text converted to KiCad footprints to make a nice card on some 0.6mm PCBs. The paper on that photo is covering the link to the build instructions, which weren't written at the time (they're *mostly* done now, I may update this post with the link later).

Finally, while I was out travelling at the start of April my new (to me) HP 4395A arrived so I've finally got some spectrum output. The output is very similar between the original and my version, with the major notable difference being that my version is 10dB worse at the third harmonic. I lack the equipment (and understanding) to properly measure phase noise, but if anyone in AU/NZ wants to volunteer their time & equipment for an afternoon I'd love an excuse for a field trip.



Spectrum with input sourced from my house rubidium (natively a 5MHz unit) via my 8140 line. Note that despite saying "ExtRef" the analyzer is synced to its internal 10811 (which is an optional unit, and uses an external jumper, hence the display note.



Spectrum with input sourced from the analyzer's own 10811, and power from the DC bias generator also from the analyzer.


1: Or at least I didn't think they had, I've since found out that there was a multi output pod, and one is currently in the post heading to me.
2: An option on the standard Spectracom pods, albeit a rare one.
southpark cartoon

Transport security for BGP, AKA BGP-STARTTLS, a proposal

Several days ago, inspired in part by an old work mail thread being resurrected I sent this image as a tweet captioned "The state of BGP transport security.":



The context of the image for those not familiar with it is this image about noSQL databases.

This triggered a bunch of discussion, with multiple people saying "so how would *you* do it", and we'll get to that (or for the TL;DR skip to the bottom), but first some background.

The tweet is a reference to the BGP protocol the Internet uses to exchange routing data between (and within) networks. This protocol (unless inside a separate container) is never encrypted, and can only be authenticated (in practice) by a TCP option known as TCP-MD5 (standardised in RFC2385). The BGP protocol itself has no native encryption or authentication support. Since routing data can often be inferred by the packets going across a link anyway, this has lead to this not being a priority to fix.

Transport authentication & encryption is a distinct issue from validation of the routing data transported by BGP, an area already being worked on by the various RPKI projects, eventually transport authentication may be able to benefit from some of the work done by those groups.

TCP-MD5 is quite limited, and while generally supported by all major BGP implementations it has one major limitation that makes it particularly annoying, in that it takes a single key, making key migration difficult (and in many otherwise sensible topologies, impossible without impact). Being a TCP option is also a pain, increasing fragility.

At the time of its introduction TCP-MD5 gave two main benefits the first was to have some basic authentication beyond the basic protocol (for which the closest element in the protocol is the validation of peer-as in the OPEN message, and a mismatch will helpfully tell you who the far side was looking for), plus making it harder to interfere with the TCP session, which on many of the TCP implementations of the day was easier than it should have been. Time, however has marched on, and protection against session interference from non-MITM is no longer needed, the major silent MITM case of Internet Exchanges using hubs is long obsolete, plus, in part due to the pain associated in changing keys many networks have a "default" key they will use when turning up a peering session, these keys are often so well known for major networks that they've often been shared on public mailing lists, eliminating what little security benefit TCP-MD5 still brings.

This has been known to be a problem for many years, and the proposed replacement TCP-AO (The TCP Authentication Option) was standardised in 2010 as RFC5925, however, to the best of my knowledge eight years later no mainstream BGP implementation supports it, and as it too is a TCP option, not only does it still has many of the downsides of MD5, but major OS kernels are much less likely to implement new ones (indeed, an OS TCP maintainer commenting such on the thread I mentioned above is what kicked off my thinking).

TCP, the wire format, is in practice unchangeable. This is one of the major reasons for QUIC, the TCP replacement protocol soon to be standardised as HTTP/3, so for universal deployment any proposal that requires changes to TCP is a non-starter.

Any solution must be implementable & deployable.
  • Implementable - BGP implementations must be able to implement it, and do so correctly, ideally with a minimum of effort.
  • Deployable - Networks need to be able to deploy it, when authentication issues occur error messages should be no worse than with standard BGP (this is an area most TCP-MD5 implementations fail at, of those I've used JunOS is a notable exception, Linux required kernel changes for it to even be *possible* to debug)


Ideally any security-critical code should already exist in a standardised form, with multiple widely-used implementations.

Fortunately for us, that exists in the form of TLS. IPSEC, while it exists, fails the deployable tests, as almost anyone who's ever had the misfortune of trying to get IPSEC working between different organisations using different implementations can attest, sure it can usually be made to work, but nowhere near as easily as TLS.

Discussions about the use of TLS for this purpose have happened before, but always quickly run into the problem of how certificates for this should be managed, and that is still an open problem, potentially the work on RPKI may eventually provide a solution here, but until that time we have a workable alternative in the form of TLS-PSK (first standardised in RFC4279), a set of TLS modes that allow the use of pre-shared keys instead of certificates (for those wondering, not only does this still exist in TLS1.3 it's in a better form). For a variety of reasons, not the least the lack of real-time clocks in many routers that may not be able to reach an NTP server until BGP is established, PSK modes are still more deployable than certificate verification today. One key benefit for TLS-PSK is it supports multiple keys to allow migration to a new key in a significantly less impactful manner.

The most obvious way to support BGP-in-TLS would simply be to listen on a new port (as is done for HTTPS for example), however there's a few reasons why I don't think such a method is deployable for BGP, primarily due to the need to update control-plane ACLs, a task that in large networks is often distant from the folk working with BGP, and in small networks may not be understood by any current staff (a situation not dissimilar to the state of TCP). Another option would simply be to use protocol multiplexing and do a TLS negotiation if a TLS hello is received, or unencrypted BGP for a BGP OPEN, this would violate the general "principal of least astonishment", and would be harder for operators to debug.

Instead I propose a design similar to that used by SMTP (where it is known as STARTTLS), during early protocol negotiation support for TLS is signalled using a zero-length capability in the BGP OPEN, the endpoints do a TLS negotiation, and then the base protocol continues inside the new TLS tunnel. Since this negotiation happens during the BGP OPEN, it does mean that other data included in the OPEN leaks. Primarily this is the ASN, but also the various other capabilities supported by the implementation (which could identify the implementation), I suggest that if TLS is required information in the initial OPEN not be validated, and standard reserved ASN be sent instead, and any other capabilities not strictly required not sent, with a fresh OPEN containing all normal information sent inside the TLS session.

Migration from TCP-MD5 is key point, however not one I can find any good answer for. Some implementations already allow TCP-MD5 to be optional, and that would allow an easy upgrade, however such support is rare, and unlikely to be more widely supported.

On that topic, allowing TLS to be optional in a consistent manner is particularly helpful, and something that I believe SHOULD be supported to allow cases like currently unauthenticated public peering sessions to be migrated to TLS with minimal impact. Allowing this does open the possibility of a downgrade attack, and make more plausible attacks causing state machine confusions (implementation believes it's talking on a TLS-secured session when it isn't).

What do we lose from TCP-MD5? Some performance, whilst this is not likely to be an issue for most situations, it is likely not an option for anyone still trying to run IPv4 full tables on a Cisco Catalyst 6500 with Sup720. We do also lose the TCP manipulation prevention aspects, however these days those are (IMO) of minimal value in practice. There's also the costs of implementations needing to include a TLS implementations, and whilst nearly every system will have one (at the very least for SSH) it may not already be linked to the routing protocol implementation.

Lastly, my apologies to anyone who has proposed this before, but my neither I nor my early reviewers were aware of such a proposal. Should such a proposal already exist, meeting the goals of implementable & deployable it may be sensible to pick that up instead.

The IETF is often said to work on "rough consensus and running code", for this proposal here's what I believe a minimal *actual* demonstration of consensus with code would be:
  • Two BGP implementations, not derived from the same source.
  • Using two TLS implementations, not derived from the same source.
  • Running on two kernels (at the very least, Linux & FreeBSD)


The TL;DR version:
  • Using a zero-length BGP capability in the BGP OPEN message implementations advertise their support for TLS
    • TLS version MUST be at least 1.3
    • If TLS is required, the AS field in the OPEN MAY be set to a well-known value to prevent information leakage, and other capabilities MAY be removed, however implementations MUST NOT require the TLS capability be the first, last or only capability in the OPEN
    • If TLS is optional, which MUST NOT be default behaviour), the OPEN MUST be (other than the capability) be the same as a session configured for no encryption
  • After the TCP client receives a conformation of TLS support from the TCP server's OPEN message, a TLS handshake begins
    • To make this deployable TLS-PSK MUST be supported, although exact configuration is TBD.
    • Authentication-only variants of TLS (ex RFC4785) REALLY SHOULD NOT be supported.
    • Standard certificate-based verification MAY be supported, and if supported MUST validate use client certificates, validating both. However, how roots of trust would work for this has not been investigated.
  • Once the TCP handshake completes the BGP state starts over with the client sending a new OPEN
    • Signalling the TLS capability in this OPEN is invalid and MUST be rejected
  • (From here, everything is unchanged from normal BGP)


Magic numbers for development:
  • Capability: (to be referred to as EXPERIMENTAL-STARTTLS) 219
  • ASN (for avoiding data leaks in OPEN messages): 123456
    • Yes this means also sending 4-byte capability. Every implementation that might possibly implement this already supports 4-byte ASNs.


The key words "MUST (BUT WE KNOW YOU WON'T)", "SHOULD CONSIDER", "REALLY SHOULD NOT", "OUGHT TO", "WOULD PROBABLY", "MAY WISH TO", "COULD", "POSSIBLE", and "MIGHT" in this document are to be interpreted as described in RFC 6919.

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC2119.
southpark cartoon

Custom output pods for the Standard Research CG635 Clock Generator

As part of my previously mentioned side project the ability to replace crystal oscillators in a circuit with a higher quality frequency reference is really handy, to let me eliminate a bunch of uncertainty from some test setups.

A simple function generator is the classic way to handle this, although if you need square wave output it quickly gets hard to find options, with arbitrary waveform generators (essentially just DACs) the common option. If you can get away with just sine wave output an RF synthesizer is the other main option.

While researching these I discovered the CG635 Clock Generator from Stanford Research, and some time later picked one of these up used.

As well as being a nice square wave generator at arbitrary voltages these also have another set of outputs on the rear of the unit on an 8p8c (RJ45) connector, in both RS422 (for lower frequencies) and LVDS (full range) formats, as well as some power rails to allow a variety of less common output formats.

All I needed was 1.8v LVCMOS output, and could get that from the front panel output, but I'd then need a coax tail on my boards, as well as potentially running into voltage rail issues so I wanted to use the pod output instead. Unfortunately none of the pods available from Stanford Research do LVCMOS output, so I'd have to make my own, which I did.

The key chip in my custom pod is the TI SN65LVDS4, a 1.8v capable single channel LVDS reciever that operates at the frequencies I need. The only downside is this chip is only available in a single form factor, a 1.5mm x 2mm 10 pin UQFN, which is far too small to hand solder with an iron. The rest of the circuit is just some LED indicators to signal status.


Here's a rendering of the board from KiCad.

Normally "not hand solderable" for me has meant getting the board assembled, however my normal assembly house doesn't offer custom PCB finishes, and I wanted these to have white solder mask with black silkscreen as a nice UX when I go to use them, so instead I decided to try my hand at skillet reflow as it's a nice option given the space I've got in my tiny apartment (the classic tutorial on this from SparkFun is a good read if you're interested). Instead of just a simple plate used for cooking you can now buy hot plates with what are essentially just soldering iron temperature controllers, sold as pre-heaters making it easier to come close to a normal soldering profile.

Sadly, actually acquiring the hot plate turned into a bit of a mess, the first one I ordered in May never turned up, and it wasn't until mid-July that one arrived from a different supplier.

Because of the aforementioned lack of space instead of using stencils I simply hand-applied (leaded) paste, without even an assist tool (which I probably will acquire for next time), then hand-mounted the components, and dropped them on the plate to reflow. I had one resistor turn 90 degrees, and a few bridges from excessive paste, but for a first attempt I was really happy.


Here's a photo of the first two just after being taken off the hot plate.

Once the reflow was complete it was time to start testing, and this was where I ran into my biggest problems.

The two big problems were with the power supply I was using, and with my oscilloscope.

The power supply (A Keithley 228 Voltage/Current source) is from the 80's (Keithley's "BROWN" era), and while it has nice specs, doesn't have the most obvious UI. Somehow I'd set it to limit at 0ma output current, and if you're not looking at the segment lights it's easy to miss. At work I have an EEZ H24005 which also resets the current limit to zero on clear, however it's much more obvious when limiting, and a power supply with that level of UX is now on my "to buy" list.

The issues with my scope were much simpler. Currently I only have an old Rigol DS1052E scope, and while it works fine it is a bit of a pain to use, but ultimately I made a very simple mistake while testing. I was feeding in a trigger signal direct from the CG635's front outputs, and couldn't figure out why the generator was putting out such a high voltage (implausibly so). To cut the story short, I'd simply forgotten that the scope was set for use with 10x probes, and once I realised that everything made rather more sense. An oscilloscope with auto-detection for 10x probes, as well as a bunch of other features I want in a scope (much bigger screen for one), has now been ordered, but won't arrive for a while yet.

Ultimately the boards work fine, but until the new scope arrives I can't determine signal quality of them, but at least they're ready for when I'll need them, which is great for flow.
southpark cartoon

Custom uBlox GPSDO board

For the next part of my ongoing project I needed to test the GPS reciever I'm using, a uBlox LEA-M8F (M8 series chip, LEA form factor, and with frequency outputs). Since the native 30.72MHz oscillator is useless for me I'm using an external TCVCXO (temperature compensated, voltage controlled oscillator) for now, with the DAC & reference needed to discipline the oscillator based on GPS. If uBlox would sell me the frequency version of the chip on its own that would be ideal, but they don't sell to small customers.

Here's a (rather modified) board sitting on top of an Efratom FRK rubidium standard that I'm going to mount to make a (temporary) home standard (that deserves a post of its own). To give a sense of scale the silver connector at the top of the board is a micro-USB socket.



Although a very simple board I had a mess of problems once again, both in construction and in component selection.

Unlike the PoE board from the previous post I didn't have this board manufactured. This was for two main reasons, first, the uBlox module isn't available from Digikey, so I'd still need to mount it by hand. The second, to fit all the components this board has a much greater area, and since the assembly house I use charges by board area (regardless of the number or density of components) this would have cost several hundred dollars. In the end, this might actually have been the sensible way to go.

By chance I'd picked up a new soldering iron at the same time these boards arrived, a Hakko FX-951 knock-off and gave it a try. Whilst probably an improvement over my old Hakko FX-888 it's not a great iron, especially with the knife tip it came with, and certainly nowhere near as nice to use as the JBC CD-B (I think that's the model) we have in the office lab. It is good enough that I'm probably going to buy a genuine Hakko FM-203 with an FM-2032 precision tool for the second port.

The big problem I had hand-soldering the boards was bridges on several of the components. Not just the tiny (0.65mm pitch, actually the *second largest* of eight packages for that chip) SC70 footprint of the PPS buffer, but also the much more generous 1.1mm pitch of the uBlox module. Luckily solder wick fixed most cases, plus one where I pulled the buffer and soldered a new one more carefully.

With components, once again I made several errors:
  • I ended up buying the wrong USB connectors for the footprint I chose (the same thing happened with the first run of USB-C modules I did in 2016), and while I could bodge them into use easily enough there wasn't enough mechanical retention so I ended up ripping one connector off the board. I ordered some correct ones, but because I wasn't able to wick all solder off the pads they don't attach as strongly as they should, and whilst less fragile, are hardly what I'd call solid.
  • The surface mount GPS antenna (Taoglas AP.10H.01 visible in this tweet) I used was 11dB higher gain than the antenna I'd tested with the devkit, I never managed to get it to lock while connected to the board, although once on a cable it did work ok. To allow easier testing, in the end I removed the antenna and bodged on an SMA connector for easy testing.
  • When selecting the buffer I accidentally chose one with an open-drain output, I'd meant to use one with a push-pull output. This took quite a silly long time for me to realise what mistake I'd made. Compounding this, the buffer is on the 1PPS line, which only strobes while locked to GPS, however my apartment is a concrete box, with what GPS signal I can get inside only available in my bedroom, and my oscilloscope is in my lab, so I couldn't demonstrate the issue live, and had to inject test signals. Luckily a push-pull is available in the same footprint, and a quick hot-air aided swap later (once parts arrived from Digikey) it was fixed.

Lessons learnt:
  • Yes I can solder down to ~0.5mm pitch, but not reliably.
  • More test points on dev boards, particularly all voltage rails, and notable signals not otherwise exposed.
  • Flux is magic, you probably aren't using enough.

Although I've confirmed all basic functions of the board work, including GPS locking, PPS (quick video of the PPS signal LED), and frequency output, I've still not yet tested the native serial ports and frequency stability from the oscillator. Living in an urban canyon makes such testing a pain.

Eventually I might also test moving the oscillator, DAC & reference into a mini oven to see if a custom OCXO would be any better, if small & well insulated enough the power cost of an oven shouldn't be a problem.

Also as you'll see if you look at the tweets, I really should have posted this almost a month ago, however I finished fixing the board just before heading off to California for a work trip, and whilst I meant to write this post during the trip, it's not until I've been back for more than a week that I've gotten to it. I find it extremely easy to let myself be distracted from side projects, particularly since I'm in a busy period at $ORK at the moment.
Error running style: S2TIMEOUT: Timeout: 4, URL: laptop006.livejournal.com/ at /home/lj/src/s2/S2.pm line 531.