All posts by Mike Clements

Ubuntu 24!

Introduction

My home desktop has been running Ubuntu 18 since it came out in April 2018. As all even numbered versions, it has LTS (long term support) for 5 years, which ended last year. Yet the reliability of Ubuntu is a double edged sword. It runs so long, without filling the hard drive with garbage, without fragmenting the filesystem, without losing performance, that over the years you customize it so many ways that by the time you need to upgrade, it’s hard to remember everything you’ll need to set up in a new system.

So why upgrade at all? After 5 years there are no more updates. No more security patches, and for most of the applications you can only run old versions. So you need to upgrade to use the latest versions of applications and stay on top of security.

You can upgrade in place to keep everything you’ve configured, with do-release-upgrade. But this only takes you up 1 step to the next LTS version (18 to 20, 20 to 22, etc.). So I’d have to do that 3 times! And it doesn’t guarantee that everything you set up will work – though it usually does.

This time around, 6 years and 3 versions behind, I decided to do a full upgrade from 18 to 24 and wipe the system boot drive. I waited until Ubuntu 24 was released on April 25 and then hit the keyboard.

Install

I actually use XUbuntu, a variant of Ubuntu that uses the XFCE desktop instead of Unity. I prefer the XFCE desktop, as it is more traditional and faster, using less CPU and RAM. Even if the computer is fast enough to support a more heavyweight desktop, why waste the CPU and RAM on that? Even if Unity was as lightweight and fast as XFCE, I would still prefer XFCE.

First I backed up my system using deja-dup. Most of my data files are on separate hard drives, which I would not be erasing. So I only needed to back up my home directory and an “apps” directory that I use for software that is manually installed. I also reviewed my apps and files to ensure that all the software I wanted to install was available on my data drive: DEB, TAR, etc. As my desktop has 4 drives, I wrote down the device and UUID for each so I could recognize them during installation.

Next I downloaded the ISO file, burned it to a DVD, rebooted to the DVD which started XUbuntu. I selected to do a manual install, reformatting the system drive. I selected “minimal installation” because I didn’t want a bazillion crapplets that I’ll never use, but also “include 3rd party and binary software”. This all went smoothly and it also automatically installed the NVidia drivers for my GTX1660.

Next, I removed the DVD and booted, my computer came up in XUbuntu 24. Yay! But now the real work begins…

Configuration

Firefox

Firefox is a mess in Ubuntu 24. First, by default it installs as a Snap, which nobody likes. And, on top of this, they’ve crippled its access to the filesystem. You can remove it from Snap with snap remove firefox and then install with apt install firefox, but the next time you update it will revert back to the Snap install. To get it to stay this way you’ve got to edit some config files to tell the system to always install it from apt.

Next, I found that Firefox could not download files. After scratching my head for a while I realized this is because my home Downloads folder is a link that points to my data disk. And I also discovered that I could not open local documentation HTML files for the Android and Java SDK. Putting 2 and 2 together, I realized that someone decided to cripple Firefox’s access to the filesystem. WTF!?

I suspect this is the “younger generation” of developers, who think the local filesystem is somehow “unsafe”, and really unnecessary anyway since they do everything in the cloud. That is the only reason I can imagine for the crazy idea of crippling such an essential function. Most sites say that installing Firefox from apt instead of Snap solves this. Perhaps it once did, but not anymore in Ubuntu 24. With some Googling I learned they’ve created an AppArmor config that restricts filesystem access for Firefox. My fix was to configure AppArmor to disable the Firefox controls, like this:

sudo ln -s /etc/apparmor.d/usr.bin.firefox /etc/apparmor.d/disable

Altogether, this got Firefox working properly again.

Backups: Deja-Dup and Duplicity

To restore all of my backups I’d have to be root because some of the files to be restored are system, owned by root not by me. No problem, right? Think again. My backups are on a separate disk that I mount to a local folder /dsk3. When I ran deja-dup to restore a backup, it would unmount that drive!

If I ran deja-dup as myself, it would work only because it wasn’t root so its attempt to unmount the drive would fail. This was fine, it still worked, even if only by accident. But with deja-dup it is impossible to restore a backup that you have mounted to a separate disk, if you run it as root.

I could not find a solution to this, so I only restored my home and apps directories mentioned above. That’s all I really needed to back up, so I got lucky.

Thunderbird

I installed from apt and it saw all of my email that was restored from the backup – both my local folders and all of my IMAP email server accounts. Bang!

Python and Virtualenv

In Ubuntu 24 Python works differently. The system Python is now 3.12 instead of 2.7, which is nice. But the procedure for installing virtualenv and virtualenvwrapper that I used for Ubuntu 22, does not work in 24.

First, use pipx instead of pip. That is, pipx install virtualenv.  After that, pipx install virtualenvwrapper. At this point they are installed but the don’t work. Next, I had to change the environment variables and python & wrapper script locations:

export WORKON_HOME=$HOME/.virtualenvs
export VIRTUALENVBIN=$HOME/.local/share/pipx/venvs/virtualenvwrapper/bin
export VIRTUALENVWRAPPER_PYTHON=$VIRTUALENVBIN/python
source $VIRTUALENVBIN/virtualenvwrapper.sh

I created some virtualenvs, listed them, etc. it all works fine – great!

Pulseaudio and Pipewire

Ubuntu 24 audio is working, but it runs pipewire instead of pulseaudio. I had pulseaudio configured to avoid resampling for bit perfect audio. Pipewire is not doing this; it resamples everything to the system default sample rate of 48 kHz. With some reading, I found a simple way to configure this.

The pipewire config files reside here: /usr/share/pipewire

In particular, the primary config file is: /usr/share/pipewire/pipewire.conf

But rather than edit this directly, you can leave it in place there and create another config file that overrides individual settings.

Create this directory if it doesn’t already exist: /etc/pipewire/pipewire.conf.d

And in that directory, create text files (having any name) that override individual settings. For example I created this file: /etc/pipewire/pipewire.conf.d/pipewire-custom.conf

# Daemon config file for PipeWire version "1.0.5" #
#
# Copy and edit this file in /etc/pipewire for system-wide changes
# or in ~/.config/pipewire for local changes.
#
# It is also possible to place a file with an updated section in
# /etc/pipewire/pipewire.conf.d/ for system-wide changes or in
# ~/.config/pipewire/pipewire.conf.d/ for local changes.
#

context.properties = {
    ## Properties for the DSP configuration.
    default.clock.rate = 48000
    default.clock.allowed-rates = [ 44100 48000 88200 96000 176400 192000 ]
}

stream.properties = {
    resample.quality = 10
}

After doing this, when I play audio files using VLC, with an external DAC connected to my sound card’s SPDIF output, the external DAC shows the sample rate changing to the native sample rate of each file that I play. It works!

Also: the DSP library PulseEffects was replaced with EasyEffects, a newer version that works with PipeWire. This new version has a feature I like: it does not resample all audio to the system sample rate like PulseEffects did, but plays audio at its native rate without resampling, changing the system rate to match the audio. Very nice!

Audio Update

Audio stopped working on each bootup. After some troubleshooting I discovered that the kernel module for my Juli@ sound card is not loaded. How to check this?

First, run lspci -k and it shows you all your PCI cards, and for each, which kernel module it uses. This tole me the card is plugged in and the module is snd_ice1724.

Next, check whether that module is loaded, or really check for all sound modules by running: lsmod | grep snd_. This did not list the snd_ice1724 module.

So, load the module manually by typing: sudo modprobe snd_ice1724. After that, repeat the above command to see if it’s loaded. Bingo! It was. And after doing this, audio started working.

Now here’s the mystery: normally you can tell the system to load new kernel modules by creating files in files in /etc/modules.load.d. For example I created this file:

-rw-r--r-- 1 root root 53 Apr 29 18:02 /etc/modules-load.d/snd_julia.conf

The contents are simple:

# kernel module for the Juli@ sound card
snd_ice1724

This should tell Linux to load the module at boot, so I don’t have to load it manually. But my system still doesn’t load this module. Or maybe it’s trying but failing, but I can’t find any errors. And I also looked for it in blacklists, but it is not a blacklisted module. Indeed, if it were the system shouldn’t let me load it manually.

At least I can fix the problem, but I don’t know why it doesn’t load automatically when it detects the sound card, nor manually from the config file, so I’m stuck loading this module manually at the command-line.

Java, Tomcat and PostgreSQL

I installed openJDK versions 8, 11 and 21, using apt, then ran update-alternatives to set the default to JDK 8. Even though it’s old, a lot of software requires it and won’t run on the newer JDKs.

Next I installed Tomcat, version 10 is what comes with Ubuntu 24. That was simple enough.

Then I needed the PostgreSQL JDBC driver so my Tomcat server could talk to databases on other servers in my home network. I found the JAR file online, copied it to /usr/share/java and linked it to /usr/share/tomcat10/lib.

Finally, I copied my audio recordings JSP app to Tomcat. First I made a directory for the app: /var/lib/tomcat10/webapps/audiorec. Then I copied the source files (JSP) to that directory. BANG! It just worked.

Apps and Games

I installed Steam and after configuring it, it saw all of my games on my separate disk drive. And it saw my configs since I restored my home directory. All good? Nope.

Two games I’ve been playing recently are Valheim and Bioshock Infinite. Steam showed how long I had been playing and when I most recently played. And the games are supposed to save your progress in the cloud. But when I started the games, they started from scratch. They had no notion of any prior game play or saves. I lost all my progress in the games.

On the brighter side, the games now run natively in Linux, where previously with Ubuntu 18 they only ran reliably from Proton, the Windows emulator. Same hardware, same drivers, the only difference is Ubuntu 24 versus 18. So I don’t know why, but that is nice…

Printers and Virtualbox

Back in 2019 I bought the TurboPrint driver for my Epson SureColor P400 printer. This is a good thing, since Epson doesn’t have a Linux driver for this printer – even though they do for most of their other models. Yet the TurboPrint driver installed and worked just fine, it doesn’t need the Epson driver. Turboprint is excellent, and essential for anyone who wants to print photo quality from Linux.

I have a Windows 10 VM for software that doesn’t run on Linux. Just a couple of apps, Epson’s CD print and Suunto’s link for my GPS wristwatch. I installed virtualbox simply from apt and it just worked – my VM came up just fine. Of course I did have to edit my system groups to ensure that I am in both plugdev and vboxusers in order for the Windows VM to be able to see USB devices like the printer and the watch.

Addendum: IOMMU

What the heck is IOMMU, you might ask? Indeed, I asked the same. I noticed a bunch of errors in dmesg, like this:

DMAR: DRHD: handling fault status reg 3

With some Googling I found it is related to a CPU chip function related to RAM called IOMMU. I’ve been using Linux since Ubuntu 8 in 2009 on a wide variety of desktops and laptops, and I’ve never seen this before. Maybe this desktop’s CPU is too old to support it – it’s an i7-4770 from around 10 years ago. Still reliable & fast, so why replace it?

The solution was to add a kernel boot parameter iommu=soft. I did this by editing the /etc/default/grub file, then running update-grub. Confirmed fixed.

Note that if I set this parameter to iommu=off, the USB system doesn’t work. No keyboard, no mouse, connect via SSH and lsusb reports no devices.

Conclusion

Of course there was a bunch of other stuff. And I’m not even done yet. But the computer is mostly working again, with most of my data and apps restored. I’ll get the rest of it working over time.

What to do with $1000

I’ve read some articles on this recently and most of them miss the most important points.

The situation: typical middle class worker who is making ends meet, but maybe not by much. You unexpectedly get $1000. What’s the best way to use this money? Let’s assume you don’t have to pay income tax on it – if you did, set aside 20% to 25% of it now.

Number 1 is to pay off credit card and similar debt. If you don’t have any, congratulations, move on to Number 2. This has high interest rates and is not tax deductible. Every dollar you put toward that pays for itself in spades. And most importantly, after paying off that debt, don’t take out more! The first step to get yourself out of a hole of debt is to stop digging. That means work out a budget within your means and stick to it, stop spending more than you earn.

Number 2 is to pay off other kinds of debt. For example, car loans. If you can’t pay it off in full, send extra money to reduce the amount of debt. However, don’t pay off a mortgage early if the interest rate is less than 5% (especially if tax deductible). Every dollar you put toward paying off a mortgage early gives you better returns if you invest it.

Finally, if you have no debt other than mortgage, invest whatever is left over. For retirement investing, put more money into your 401K or IRA. If it’s already maxed out, you can put after-tax dollars into a Roth IRA. Even if you make too much money to quality for Roth, you can still do it through the “Roth backdoor”. That is, put after-tax dollars into a traditional IRA, then immediately roll them over into a Roth IRA. For long term investments (more than 5 years), equities and equity funds are the way to go. For 1-5 years, reduce volatility with a blend of stocks and fixed income. For < 1 year, fixed income like a CD is best.

Audio Phase: Shift versus Inversion

It is said that a 180* phase shift is the same as a polarity inversion. That is, it flips a wave to its mirror-image across the time axis. If we imagine a simple sin or cos wave, we see that this is true. 180* is half a wavelength, slide it that distance either forward or back, and you get the same wave with polarity inverted. Another consequence of this lies in audio room tuning. If the distance between 2 walls is half a wavelength of a particular frequency, the wave reflecting from the wall, being inverted polarity, cancels the wave arriving which causes a dip or null at that frequency. Those same walls will also boost waves at twice that frequency because that same distance between the walls is their full wavelength, so the reflected wave is in phase with the arriving one.

But this doesn’t work with a general musical waveform. No amount of sliding it left (back) or right (forward) in time will invert its polarity. Intuitively, we see that a musical wave is not symmetric or repeating like a sin or cos wave. The musical waveform is much more complex, containing hundreds of frequencies all superimposed. Any distance we slide it left or right represents a phase shift at only 1 particular frequency. Alternately, sliding it left or right can be seen as a phase shift at all frequencies, but a different phase angle for each, since the distance it shifted is a different number of wavelengths for each frequency it contains. As in the above example, it boosts some frequencies and cuts others. This is what happens in a comb filter.

Since every frequency has a different wavelength, it’s hard to imagine how a phase shift of the same angle at all frequencies could even be possible. It is possible, but to do it we need to expand into another dimension and use complex numbers. That computation creates a new waveform that is the polarity inverted version of the original. You can find explanations of this all over the internet, for example here: https://www.audiosciencereview.com/forum/index.php?threads/analytical-analysis-polarity-vs-phase.29331/

Because of this, when speaking of music and audio I prefer the term “polarity inversion” to “180* phase shift”. Even though they can be equivalent, the former is concise while the latter is somewhat ambiguous since one must also specify at what frequencies the phase shift is applied.

Email Send, IP Addresses, Blacklisting

About a month ago emails that we sent were bouncing, being rejected by the destination servers saying one of the IP address where it came from was blacklisted. I host this site and our email through Hostgator, using their SMTP server to send email. So my first guess was that the Hostgator SMTP server was blacklisted. It’s plausible, as it’s shared by many customers, some of whom may be spammers. But the IP address from the email bounce message didn’t match that server. Turns out it was the IP address that Comcast assigned to my home router. So Comcast assigned me a blacklisted IP address! Perhaps Comcast shares that IP with other customers and one of them has been spamming.

However, email sent from my GMail SMTP was not bouncing. Looking at the email headers, GMail’s SMTP does not forward the IP address of the client who sends it; it sends its own. So one possible solution would be to get the Hostgator SMTP to stop forwarding my IP address, but instead use its own, or even use my own IP address for this domain, mclements.net. I contacted Hostgator support and they said they could not configure the SMTP server to do this.

In the meantime, as a temporary workaround I configured my email send to use the GMail SMTP and CC my other email. It’s clumsy but works.

My next option was to change my Comcast IP address. I called support and they confirmed that Comcast uses DHCP, so getting a new IP address should be as simple as turning off my modem & router, then turning them back on. I knew it wasn’t that simple because I had done that and the IP address never changed. The DHCP has a lease with a time duration. You need to disconnect long enough for that to expire before getting a new IP address. I left them off overnight and still got the same IP address.

DHCP servers often (though not always) use the client MAC address as a hash when providing IP addresses. If you can change the MAC address of the router that connects to your cable modem you are likely to get a new IP address. I dug into my router menus and found a config option to do exactly this: you can enter any MAC address you want, or have it copy the one from the PC you are using to connect to the router. When setting MAC addresses manually like this, one must be careful because the MAC address is not just a random number. The first several hex codes are assigned to manufacturer and device type. You should ensure that what you enter is a valid MAC address.

In short, this worked. After changing my router’s MAC, Comcast issued a different IP address that was not blacklisted, and now my email is not bouncing. Since these IP addresses belong to Comcast, I called to let them know the old IP address was blacklisted so they can deal with it.

Summary of steps:

  1. Configure your router to use a different MAC address
  2. Turn off the cable modem and router
  3. Turn on the cable modem and let it boot up and sync
  4. Turn on the router

After step (4) it may take longer than usual to get an internet connection. When the router negotiates with Comcast for a connection, if its MAC address is different and Comcast assigns a new IP address, the process can take longer.

Bikes: Electronic Shifting

In late 2001 I bought a new road bike, a Fezzari Empire. One of the reasons I bought it at that time is because that was the last year that Shimano offered its best components (Ultegra and Dura Ace) with mechanical shifting. Since then, they are only available with electronic. SRAM has also gone to the “dark side”.

Why don’t I like electronic shifting? Everyone else seems to love it. They say it works perfectly. Perhaps it does (when it’s not failing due to dead batteries) but mechanical shifting works perfectly too. It’s been perfected over decades and it is simpler, cheaper, lower maintenance, more reliable and durable.

My reasons include:

Electronic shifting has batteries that can die on a ride. To be safe, you need to add a spare battery to your ride kit. These batteries also must be charged periodically, and replaced when they wear out.

Mechanical shifting, once adjusted properly, works perfectly for several years and thousands of miles without needing adjustment or maintenance (other than periodic cleaning, which electronic shifting also requires).

Electronic shifting is less durable and reliable. Consider a multi-day stage ride. Electronic shifting requires you to bring extra batteries or find a way to charge them. It’s extra hassle with no corresponding benefit. Lachlan Morton, who won the Tour Divide in 2023, had electronic shifting. The batteries were a hassle during the ride (charging and replacing), and later it failed, forcing him to jerry-rig his derailleur with a spare spoke.

Electronic shifting is also considerably more expensive.

Electronic shifting has no real benefit to justify the expense and hassle. It’s not lighter, nor more efficient, nor faster, and it’s actually less durable and reliable.

Getting philosophical, electronic shifting contradicts the classic purity of what a bicycle should be: a simple, elegant, purely mechanical machine. No batteries, no electronics, no software. The only power source should be the person riding it.

In short, electronic shifting is a complex, expensive, fragile solution to a problem that doesn’t exist.

So why do so many riders like it? It’s high tech, more expensive making it more exclusive, and the latest trendy shiny object. Pros use it, which can make sense since their bike only has to last 1 day, they don’t care how much it costs or how long it will last. But for those of us who aren’t GC contenders in the Tour de France, I’m surprised so many cyclists drink the kool-aid and can’t see that the Emperor has no clothes. And bike companies love it because they can charge you 3 times the price for the equipment, lock you into their “ecosystem”, get a new revenue stream selling batteries, and every few years forcefully deprecate old systems forcing people to upgrade.

No thanks, I have mechanical, the shifting is sublime, it’s simple and user-serviceable, and it will last me the next several decades until (God forbid) I’m too old to ride bikes anymore.

Tubeless Tires: MTB Yes, Road No

MTB: Yes

Tubeless tires are great for mountain bikes. They address the 2 biggest limitations of the tube+tire technology that had been used for more than 50 years: traction and puncture resistance. On MTB trails with tubed tires, you must run high enough pressures to avoid pinch flats. And those pressures are high enough to greatly reduce traction. For example, on typical 26″ x 2.3″ tires you would typically need 50 PSI to avoid flats, which makes the tire almost as hard as the rocks you’re riding over. And that only avoids pinch flats. You would still get flats from thorns or sharp rocks.

With tubeless tires you can run about half the pressures, 25-30 PSI, on those same size tires. This is a vast improvement in traction. And despite these lower pressures, the tires are even more flat resistant than before. It really is the best of both worlds.

Tubeless has its drawbacks. It makes tire mounting messy, more labor intensive, and requires a compressor. It also requires more maintenance, as the fluid must be replaced every few months. And once or twice a year you also need to completely remove the tire from the rim to clean out all the old dried up sealant.

But when mountain biking, the benefits are are well worth these hassles.

Road: No

None of these benefits apply with road bikes. With their skinnier tires, the ideal pressures for rolling resistance, traction and comfort are already high enough that you would run the same pressures whether tubeless or not. On road bikes, pinch flats with tubes are not an issue and flats are much less common.

Yes, tubeless is more puncture resistant. But road riders don’t get punctures nearly as often as MTB riders used to. And when those punctures happen on the road, the tubeless sealant doesn’t handle them as effectively as it does with MTB tires. Several times over the years I’ve stopped to help road riders with tubeless tires fix flats when their sealant just sprayed everywhere making a mess instead of sealing the leak. They didn’t have a spare tube because they thought tubeless tires were “flat-proof”.

On my road bike I use latex inner tubes and Conti GP5000 tires. I ride it at least 5,000 miles per year, usually more. If I switched to tubeless:

  • My wheels would not be any lighter.
  • My wheels would not be any faster.
  • My tire pressures would not change.
  • I would still have to carry a spare inner tube with me.
  • If I did get a flat on a ride, it would be more of a hassle than just swapping an inner tube.
  • I would spend more time (not less) maintaining my tires because I would have to clean out & replace the sealant more often than I get flats today.
  • That maintenance would also take longer and be messier, due to dealing with sealant and a compressor.

In short, road bikes don’t need either of the 2 key benefits that tubeless tires provide, so all it does is create hassle and mess.

Tascam DA3000

The Tascam DA-3000 is a professional rack mount digital recorder. For years I owned a prior model, the SS-R1. It provided years of reliable service and I used it to archive nearly 1,000 vinyl LPs. The DA-3000 improves upon the SS-R1 in a few important ways:

  • Better AD converters: Burr Brown PCM4202
  • Better DA converters: Burr Brown PCM1795
  • Supports higher data rates: 24 bit, up to 192 kHz PCM and DSD 128 (5.6 MHz)
  • Direct AD-DA mode
  • Lower distortion and noise

It also retains many of the most important features of the SS-R1

  • Records to SD and CF media (no internal hard drive)
  • No fans – dead silent
  • The flexibility of many connections
    • Analog unbalanced RCA input and output
    • Analog balanced XLR input and output
    • SPDIF coax input and output
    • AES/EBU input and output
    • Internal or external clocks
  • Professional build quality, durability and reliability

Tascam no longer makes the DA-3000 so I bought mine used from eBay. In using it I’ve discovered some interesting quirks.

Date/Time Clock

The SS-R1 had a clock that had to be set every time you powered it on. Even when plugged in, it would forget the date & time when turned off. The DA-3000 fixed this – at least supposedly. But every time I powered mine up, I had to set the clock.

The problem was that the DA-3000 uses a rechargeable button cell battery to remember the clock when turned off. And it uses a tiny one that is soldered to the board. After a year or two, this battery dies and is not easy to replace (you must disassemble the unit, remove the board and solder). I contacted Tascam support and they said they no longer provide this service. It was annoying enough I decided to do my own permanent fix, better than what they would do at the factory.

Rather than simply replace that soldered-in battery, I installed a small battery cage for a CR2032 battery, which has the same voltage but is more than twice the size and capacity. Instead of soldering a new battery onto the board, I soldered the battery cage lead wires. Here’s what it looks like now:

I put an ML2032 battery into the cage (the rechargeable version of a CR2032). Not only will it last much longer than the tiny OEM battery, when it dies I can now replace it in 10 minutes easily without any soldering. This is how the DA-3000 should have been built from the factory.

DA-AD Mode

This mode stands for “Digital-Analog, Analog-Digital”. In this mode, the DA-3000 doesn’t record, but merely activates its DA and AD converters. You select which analog & digital inputs & outputs it uses. The DA converter has a slightly warm, soft voicing free of glare. Very nice. I can find no explanation for this in measurements, as its frequency response and distortion measure clean.

DA-AD mode does not auto-detect the sample rate. You must select the sample rate in the menus. If the sample rate you select does not match the digital input, the DA-3000 will still produce analog output but it is distorted. By this I mean high frequencies rolled off with elevated harmonic distortion.

Frequency response in DA-AD mode with sample rate mismatch, compared to sample rate match.

Distortion in DA-AD mode with sample rate mismatch:

For comparison, here’s distortion when you manually set the rate to match the input:

This distortion is measured after the DA3000 DA conversion and the SSR1 AD conversions. The distortion when you don’t manually set the sample rate to match the input is not documented in the manual, and Tascam support did not respond to my inquiry about this. So just know about this and set it!

Sample Rate Sensitivity

The popular S/PDIF digital format (whether coax or toslink) is a “push” protocol. The source device sends data to the downstream target using the source’s clock. The target device has no way to tell the source to slow down or speed up. No two clocks ever agree exactly, so the target device has to adapt to the source device sample rate. In contrast, audio over USB is a “pull” protocol. The downstream target device runs off its own clock, requesting data from the source as needed. No need to synchronize clocks or adapt sample rates.

Lest anyone disregard S/PDIF clock sync as a problem “solved in the real world”, consider that well engineered DACs use TCXO for their clock, which are temperature compensated crystal oscillators. These are typically accurate to roughly 1 ppm, which at CD quality 44.1 kHz makes a clock drift of 1 sample every 22.6 seconds. So the issue is real – with S/PDIF, the downstream device must adapt to the upstream clock. Buffering can’t entirely solve this, because it can only solve variations around an identical center frequency. Put differently, no two devices will ever agree on the center frequency, one will always be slightly slower than the other, which means any buffer you use will eventually under/over flow. With S/PDIF the downstream device must not only buffer the data, but also adapt its clock to the source rate.

The Tascam DA-3000 specs say it can sync to any input digital sample rate within a range +/- 100 ppm. This should be plenty, about 100x greater than the drift expected from a well engineered DAC. However, in my setup I have a toslink-coax converter between my Corda Soul (preamp/DAC) and the DA-3000. This converter causes quite a bit of jitter, so much that the DA-3000 occasionally loses sync. For example, a REW frequency sweep played through the Soul and captured on the DA-3000 looks like this:

That capture was at 88.2 kHz sampling, but it happens at all sample rates. My Topping E70 DAC handles this jitter just fine and is super clean, because it has a setting called DPLL that controls how much sample rate variance it can accept and adapt to. I had to bump it up a couple of notches to handle the switchbox.

Fortunately, the DA-3000 can do the same thing, even though one method for doing it is undocumented. When recording, enable SRC which is Sample Rate Conversion. This won’t actually convert the sample rate, because you’ll still manually set the DA-3000 sample rate to match the digital input. But when SRC is enabled, the DA-3000 accommodates and adapts to a wider range of jitter.

A better, cleaner method is to change the DA3000 clock setting from “Internal” to “DIN”. This tells the DA3000 to use the digital input as its own reference clock.

When you do either of these, distortion in the REW sweep is super clean like it should be:

AD Converter HF Noise

Another quirk of the DA-3000 is super high frequency noise in its AD converter. The noise is at 100 kHz, so you won’t see it at sample rates of 96k and lower since it’s above the Nyquist frequency. But at 176.4 and 192 kHz, it is there and looks like this:

So, if you are recording from analog inputs, don’t use 176.4 or 192 kHz. Use 88.2 or 96 kHz instead. The sound quality will be better! This is not an issue if you are recording digital inputs – that is just a bit perfect copy.

Analog Output Level

Normally, a recorder’s analog output level isn’t that important. But when using the DA-3000 in AD-DA mode, it becomes so. You need to match the input voltage sensitivity of the downstream preamp. The DA-3000 has a setting called “Reference Level” that sets this. Indirectly, this sets the analog output voltage for digital peak levels. The range is -9 dB which is +15 dBu (quiet) to -20 dB which is +24 dBu (loud). In Volts RMS this ranges from 4.36 to 12.28. The first is the consumer audio standard, the second is the professional standard.

Most consumer preamps have voltage sensitivity for their balanced analog inputs that expects peak levels around 4 V. Higher voltages can cause them to clip or distort. So you would set the DA3000 to -9 dB.

The Corda Soul has an internal switch to set its analog input voltage sensitivity. The default setting is for professional audio, expecting 12 Volts (low gain). Flick the switch to the other position and it changes to consumer, expecting 4 Volts (high gain). With the Soul set to low gain, the DA3000 setting that matches the output level of the Soul’s internal DAC is -16 to -18: -16 is about 0.5 dB quieter and -18 is about 1.5 dB louder.

Conclusion

The Tascam DA-3000 is a wonderful recorder. It is incredibly flexible, easy to use, with SOTA transparent sound quality and professional build quality. It has many other features not described here, since I’ve focused on its quirks. I’ve wanted one for years and I am so happy I finally found it!

Like any piece of gear, it has a few quirks as seen above. But none of these are serious problems, they all have workarounds.

Electronic Gyro Drift Correction

Introduction

A magnetic compass is still a required navigation instrument in airplanes. The most common type is called a “whisky compass”, mounted at the top center of the panel. The compass is tilted toward the pilot to make it easier to read. Yet this also makes it accurate only in straight and level flight. When turning, the compass’ balance masses and tilt make its reading lead or lag the airplane’s actual heading.

Most airplanes also have a directional gyro. The gyro’s rotational inertia keeps it in a stable position as the airplane rotates around it. This means it reads accurately when turning. But gyros slowly drift over time. This means during straight and level flight, the pilot must occasionally check the gyro and manually set it to the compass heading. How often is occasionally? Depends on the gyro. For gyros that are entirely independent having no external correction, it’s about every 15 minutes. And this is true whether it is mechanical or electronic.

Gyro Drift

Gyro drift is caused by two factors: the rotation of the Earth, and friction (for mechanical) or noise/errors (for electronic).

The Earth rotates through 360* every 24 hours, which is 15* per hour. The gyro is immobile in space independent of the Earth’s rotation. Thus as the Earth rotates, the gyro “moves” relative to the Earth, and since the Earth is our frame of reference, this causes the gyro to drift. The relative motion of the Earth can make a theoretically perfect gyro drift up to 15* per hour from the pilot’s frame of reference.

Bearing friction and electronic noise are more intuitively obvious causes of gyro drift. With electronic gyros we have the advantage of being able to apply software corrections. Electronic gyros are based on accelerometer sensors, which means the sensor readings must be mathematically integrated to get position. Integration cumulatively amplifies small sensor errors. For example, even if the sensor’s readings consistently average the correct value over time, each individual reading will be slightly more or less. And these accumulate over time into increasing errors.

Drift Correction

One form of drift correction is when the pilot sets the gyro to match the compass heading. Immediately after this we can assume the gyro’s heading is correct. If we store each of these changes, we have a history of how much the gyro has been drifting and can use that to auto-correct and reduce drift.

Correcting this automatically applies only to electronic gyros, since we need a software algorithm to compute and apply this.

Bias and Variance

Errors and noise fall into two categories: bias and variance. Bias refers to anything systematic or predictable, variance is the unpredictable random portion of the errors. We can detect and correct for bias but not for variance. We must be careful, because misinterpreting variance as bias can increase errors rather than reduce them.

The basic idea is that after each pilot correction, we compute the rate of drift of the correction and continue to apply that to the gyro going forward. For example, if the most recent correction was +10* and it was made 15 minutes after the prior correction, the correction is +0.667* per minute, so we automatically apply that to the gyro going forward.

However, it’s not quite that simple so the idea needs refinement.

For example, suppose the drift that the pilot is correcting reverses direction each time. In this case, if we correct as above, we would actually amplify rather than reduce the drift, making things worse rather than better.

The problem is that errors are a mix of bias and variance, yet our idea only works to reduce bias, not variance. One simple way to differentiate bias from variance is to look at whether recent user corrections all go in the same direction. When this happens, there is a simple linear component to the gyro errors: it’s consistently drifting in the same direction, whether clockwise or counter (this is not the only form of bias, but it’s the simplest and easiest to detect). Yet even a blind squirrel sometimes finds a nut, and random errors will sometimes also go in the same direction. When you flip a coin, you will sometimes get heads several times in a row.

Consider that with variance (completely random errors), each pilot correction is 50% likely to be in either direction, like flipping a coin. If you flip a coin twice, you get 2 heads or 2 tails half the time. Similarly, with pure variance and no bias, about half the time each pilot correction will be in the same direction as the prior correction. Three consecutive corrections in the same direction would happen about 25% of the time. Conversely, we can say that in this case the error is 75% likely to have some bias.

So we should not apply our automatic correction unless the most recent N pilot corrections were all in the same direction, and N should be at least 3. Also, we should shrink the auto-corrected rate accordingly. For example when N=3 the error is 75% likely to have some bias, but it will always have some variance too. So let’s assume that 75% of the error is bias and shrink the correction applied going forward to 75% of the pilot-entered value. In the above example, the +0.667* per minute becomes 0.5* per minute.

Oscillation and Damping

At this point we have a simple algorithm that should improve the gyro accuracy. Yet we can improve it further without adding complexity. The idea is that our method to discern bias from variance is always imperfect, and if we get it wrong it makes things worse, not better. It is better not to correct errors, than to make them worse. Put differently, if we are too aggressive with error correction we can make things worse, while if we are too passive or conservative, it still improves things just not as much.

So, we will apply a damping factor to our corrections, shrinking them just a bit. Pick a constant scaling factor between 0.0 and 1.0 and apply that to the correction. For example, suppose we pick 0.85 or 85% and N=3. With the above example, suppose the last 3 pilot corrections were all in the same direction, and the most recent one was +10*, and it was made 15 minutes after the prior one.:

  • The raw error being corrected is 10* / 15 minutes = 0.667* / minute.
  • Because N=3 we are 75% confident there is bias in this error, so shrink it to 75% of its value.
  • Apply our damping factor of 85%.
  • This makes the auto-correction factor 0.667 * .75 * .85 = 0.425* / minute
  • Apply this rate correction automatically going forward

Overall, we end up with a single pilot correction rate that is maintained in cumulative fashion. For example, in the last step above we don’t just set it to 0.425* / minute, but we add 0.425* / minute to whatever the existing value is. And repeat every time the pilot sets the gyro, so the value changes over time, adapting to varying conditions during flight.

Conclusion

This practical example is over-simplified but it illustrates the basic concepts involved regarding bias vs. variance in errors, how to differentiate them and make corrections, and how to increase our confidence that our attempt to reduce errors doesn’t unintentionally make them worse.

LineageOS – Open Source Android

Summary

Android is essentially a Linux distro. It is a mobile device oriented UI / desktop running on a version of Linux. However, it’s not really open source. Every phone manufacturer writes binary drivers to get it to run on their hardware, and they don’t contribute those to the community. And mobile carriers pile their own add-ons on top of this. So by the time the Android phone gets into the hands of the user, it is loaded with proprietary software and bloat-ware.

For example Samsung modifies Android with “TouchWiz” which significantly changes the UI. And Motorola pre-installs the Facebook app – users can “uninstall” it, but it reinstalls itself every time the phone boots. Some people might like these changes, but I think most, like me, would prefer the pure unadulterated open source Android without bloat-ware or crapplets to burn down the battery and hog the storage.

Another issue with Android is that most manufacturers only support a given model of phone for a year or two. After that, no more updates, which means the phone is condemned to planned obsolescence.

In some cases we can avoid these issues, keep our phones for years while running the latest version of open source Android without bloat-ware. If that sounds interesting, read on.

Unlocking or Jailbreaking

The term “unlocked” has 3 different meanings:

  • Carrier unlocked: you can use the phone on any carrier’s network, so long as you have the right SIM card, and the phone’s modem & radio supports the right frequencies & protocols (GSM vs. CDMA). Many phones are carrier locked when new. When buying a new phone, when it says “unlocked” it means carrier unlocked.
  • Bootloader unlocked: just about all phones are shipped with a locked bootloader. This means you cannot replace the operating system. You can only install factory supported updates using the phone’s settings menu.
  • Rooted: the phone allows the user and apps to take on “superuser” or admin privileges. This means the phone can be used as a little computer without any restrictions – direct access to the filesystem etc.

“Jailbreaking” refers to rooting Apple phones while “unlocking” is a more general term.

Carrier: Every carrier is required by law to give mobile phone users a code to carrier-unlock their phone, so long as the phone is fully paid for. Most make this process as difficult as they can in order to discourage users from doing it. But, the process is accessible to non-technical users. Essentially, they send you a long code that you enter into one of the phone’s settings menus.

Bootloader: There is no law that I know of requiring manufacturers to allow users to unlock their phone’s bootloader. Some (like Motorola) officially support this, others (like Samsung) support it only unofficially, and others do not support it at all. In the latter case, hackers often (but not always) figure out how to crack it. When you unlock the phone’s bootloader, the warranty becomes void. Of course, that’s a nothing sandwich if the phone is already out of warranty.

Rooting: This enables you to use the phone for things you can’t do otherwise. For example, direct access to the full filesystem, even system partitions, enables full backups just like a computer, and makes possible amazing customizations. However, some security apps (for example banking and 2FA) detect whether a phone is rooted and refuse to run. So don’t root unless you really need to, and take measures to handle these cases.

Booting to Recovery

Recovery is disk partition on the phone that has a mini-OS that runs as root and enables you do make changes that aren’t possible when it’s booted normally. This includes changing the partition table, wiping the system partition, loading the operating system, etc. The two most common ways to boot to recovery are:

  • A button chord: power off the phone, then power it on while holding down other buttons. Exactly which buttons varies by manufacturer. For example, with Motorola press and hold the volume down button while powering on.
  • Android tools: Android, being open source, has a full set of developer tools that are freely available to everyone. Install the Android toolkit on your computer (Linux, PC or Mac), connect your phone via USB, and control it from the computer using tools like adb and fastboot.

Every phone comes with a factory supplied recovery, but it is not intended to load custom operating systems. Two of the most popular custom recoveries are TWRP (on Samsung) and boot.img (on Motorola).

Partition Table

The partition table determines how much disk space is allocated for recovery, for the operating system, and left over for user storage. Running a newer version Android or different recovery sometimes requires more space, which means changing the partition table.

Google Apps

Google Apps includes the Google Play App Store, Contacts, Calendar and several other apps that are not part of the Android operating system, but run with special privileges other apps don’t have. These apps cannot be installed from the Play Store but are installed while booted to recovery, just after installing the operating system.

These apps are not strictly required to use the device, but without them you won’t have the Google Play store or other important functions. They are required if you want your device to work like the normal Android that everyone is familiar with.

Compiled ready to install packages of these free open source versions of Google Apps are published in two popular places: OpenGapps and MindTheGapps.

The Process

So, how does one actually do this?

  1. Find out exactly what kind of device you have, including the specific model number. For example, not just a Motorola G7 Power but model XT1955-5, because there are several different versions having different processors, radios and modems.
  2. Ensure your device is fully functional with the OEM ROM, go to settings and install all updates to ensure it has the most recent version.
  3. Ensure to actually use the features from the OEM ROM, to ensure your device is properly registered on the carrier network. For example, on T-Mobile make phone calls & texts over WiFi.
  4. Back up any data on your device that you want to keep, because the install will wipe the entire device.
  5. Find a ROM that is supported on your device. You can get an officially supported ROM, like those at lineageos. Or you can find an unofficial ROM at places like xdaforums, where individual developers create and support them. An unofficial ROM could be LineageOS or many other versions of Android.
  6. Unlock the bootloader on your device. How to do this will vary from one device to another so you’ll have to do some homework.
  7. Follow the ROM installation instructions. This typically includes these steps:
    1. Boot to the OEM recovery
    2. Flash a new recovery
    3. Boot to the new recovery
    4. Update the partition table
    5. Install the new ROM
    6. Install Google Apps
  8. Boot the device into LineageOS.

Twin Engine Airplanes

It’s common perception that twin engine airplanes are safer. Any for obvious reasons! Who wouldn’t want an extra engine? Yet the details give a more nuanced perspective.

With passenger jets, twin engines are definitely safer – no doubt about it. But with piston engine small aircraft (e.g. light twins), the safety record is more mixed. It boils down to 3 basic reasons:

  1. Having two engines doubles the likelihood of an engine failure.
  2. When one engine fails, the other produces differential thrust requiring immediate corrective action from the pilot to avoid loss of control.
  3. The single-engine performance of some piston twins is so marginal we sometimes say, “the remaining engine always has enough power to get you to the scene of the crash”.

Many people don’t really consider the first point, but when pointed out, it’s so obvious it doesn’t require further discussion.

Regarding point 2: engine failure in any airplane is an emergency, no matter how many engines it has. Yet with a single engine, the immediate pilot actions are pretty simple: keep the nose down so you don’t stall, and pitch for ideal glide speed. The engine is centered, so when it dies the airplane remains inherently stable and will keep gliding even hands-off. With twin engines, when one engine dies the other produces differential thrust that skews the airplane sideways and will flip it over if not corrected. This differential thrust is a double-whammy: the dead engine’s prop creates drag, pulling that side back, while the operating engine produces thrust, pulling the opposite side forward. Not only must the pilot keep the nose down to avoid stalling, but he must also apply heavy opposite rudder (not aileron) and feather the dead engine’s propeller to keep the airplane flying straight. If the pilot fails to do these actions quickly – within seconds – or does them incorrectly, the differential thrust can cause an uncontrollable spiral or spin.

Regarding point 3: one of the reasons people fly twins is for the superior payload and performance. You can carry heavier loads, and fly faster and higher. Yet if you are actually using that performance, you may operating in a way that cannot be supported by a single engine. So when one engine fails, even if you apply the correct inputs to keep flying, you may not be able to maintain level flight even with the good engine at full power.

Overall, the incident/accident statistics for light twins in general aviation is no better than single engine airplanes. Given this, why are big commercial jets always multi-engine for safety? Commercial aviation mitigates these factors:

  1. They use turbine engines, which are more reliable than pistons.
  2. The pilots are better trained, more frequently, and follow more strict operational limits set by both the FAA and their airline.
  3. The engines have much greater power than pistons, capable of maintaining level flight at high altitudes for extended periods of time, even when the airplane is at max weight.

In summary, light twin aircraft can be safer, or more dangerous, depending on the airplane, the pilot, and the mission or how the flight is operated. Pilots considering light twins should consider these limitations, how the airlines mitigate them, and incorporate that into their flying. For example:

  1. Maintain the aircraft above & beyond required minimums.
  2. Train yourself well beyond the required minimums, stay current.
  3. Don’t load to max weight, and fly missions that give you a healthy safety margin below the aircraft’s max performance.

Even then, in my opinion light twins are not safer and the higher performance is not worth the expense and hassles of the higher cost of fuel, maintenance, insurance and training. Speed is proportional to power cubed and drag is proportional to speed squared, so all else equal a twin burns 59% more fuel to go 26% faster.

Here’s where those figures come from:

  • The cube root of 2 is 1.26, so twice the power is 1.26 times as fast.
  • Drag is proportional to the square of speed, and 1.26 squared is 1.59.
  • Fuel consumption is proportional to drag.