The cable is there because of how guest networks are run. A guest network can isolate its clients from one another, so a laptop and a Pi that have both joined the classroom Wi-Fi may be unable to reach each other even though each one reaches the internet. A cable between the two does not depend on the network’s settings, so every step that talks to the Pi from the laptop uses the cable.

On Windows the ssh client is the OpenSSH Client, under Settings > Apps > Optional features; macOS and Linux include one.

The stand’s README has the print settings: PETG if you can, no supports, and the platter printed upside down. It turns the tripod on the array’s own center, so the source can stay put while the board rotates, and its engraved scale reads the angle.

I also keep fallback cards, copies of one finished kit. A fallback card gets a kit running when a build cannot be finished in the period; it is named for your kit before it goes in, and it still needs Steps 21 to 24, because the calibration it carries belongs to my kit.

Use the 5 V, 3 A supply in the kit, plugged into the PHASER’s USB-C input, labeled P16. The PHASER board powers the Pi and the Pluto through its own connectors, and a supply plugged into the Pi bypasses the board’s supply.

Linux writes to the card while it runs, so pulling the power or the card mid-write can corrupt it, and a corrupted card means flashing it again. Shut down with the button on the back of the PHASER, or by holding the Connected pill for two seconds, and wait until the lights go out before you unplug (Step 25).

The amplifiers behind the patches and the ADAR1000 beamformers are static-sensitive parts. Touch a grounded metal surface before you handle the board, and hold it by its edges. Keep fingers and objects off the patch face: anything on a patch changes its match, and pressure can damage it.

An SMA connector needs only finger tightness. A wrench, or a connector started at an angle, can strip the threads or break the connector off the board. Support every cable so its weight does not hang on the board’s connectors.

If the kit stops responding, run sudo reboot first, then re-run the provisioning script (Step 16), which repairs the installation without erasing the card. Flash a new card only when I tell you to.

Kits stay in the classroom. Ask me before you take one anywhere else.

Kuiper 2021_R2 is the release the Phaser software is tested on, and it is based on Debian 11, bullseye. The later Kuiper releases move to a newer Debian that changes the parts of the system the provisioning script and the Wi-Fi steps rely on: where the boot files live and which program manages the network.

Imager offers no OS customization for a custom image, only for Raspberry Pi OS, so nothing on the card changes the user and password that every later step relies on; the provisioning script in Step 16 names the kit instead.

After the write, the card holds two partitions: a small FAT one that holds the boot files, which Windows can read, and a Linux one, which it cannot. That is why Windows offers to format part of the card, and formatting it would erase the image.

The PHASER board supplies the Pi through its own connector, and powering the Pi directly bypasses the board’s supply. Unplugging the kit while Linux is running can also corrupt the card; see the last frame of Part A for the safe way to power down.

Nothing on a bare cable hands out addresses. After about a minute of asking, the Pi and the laptop each give themselves a link-local address, one in 169.254.x.x that only the cable can carry, and the laptop finds the Pi by name with multicast DNS, the .local lookup. That is why a direct cable works with no setup on either end, and why the first minute after power-on shows nothing. On my kit, hostname -I showed exactly one address at this point, 169.254.210.227.

Kuiper 2021_R2 is based on Raspberry Pi OS bullseye, where wpa_supplicant joins the Wi-Fi network and dhcpcd then requests an address on it. The Wi-Fi country is a regulatory setting: the channels a radio may use depend on where it is, so the operating system keeps the Wi-Fi radio blocked (rfkill) until a country is set. The network block names the network, and key_mgmt=NONE tells wpa_supplicant it has no password. tee -a appends the block with root privileges, which a plain >> redirection would not have.

Two things must hold on an open guest network for this step and for provisioning, and I check both before class. The network must not put a captive portal, a sign-in or terms page, in front of a new client, because the Pi has no browser to click through one. It must also allow outbound HTTPS, because the provisioning script downloads the Phaser software from GitHub.

A Raspberry Pi has no battery-backed clock, and Kuiper ships with no network time client, so a fresh card boots believing it is the day the image was made, in April 2023. Every secure download that follows checks the server’s certificate against today’s date, and a certificate is only valid between two dates, so a wrong clock makes the download fail with a certificate error before any software arrives. The provisioning script installs a time client and fixes the clock for good, but it has to be downloaded first.

The command breaks that circle with a plain, unencrypted request, which no certificate guards: curl -sI fetches only the headers of a Debian web page, sed keeps the line that begins Date:, and date -s sets the clock to it.

provision.sh turns a stock Kuiper card into a PHASER kit. In order, it checks the machine, installs a time client and fixes the clock for good, installs the system packages, adds the PHASER board’s device-tree overlay to the boot configuration, sets the hostname, installs the plumbing that lets the Pi talk to the Pluto over USB, builds pyadi-iio, ADI’s Python library for the board, and finally runs install.sh, which installs the Phaser backend and the browser interface as a service that starts at every boot.

The reboot in Step 17 is what makes the board appear. The device-tree overlay is read only at boot, and it is the file that tells Linux which chips sit on the board’s bus; until the Pi restarts with it, the operating system does not know the phase shifters exist. The new hostname also takes effect at boot.

Every kit needs its own name because every kit also joins the same Wi-Fi. Twenty kits all calling themselves phaser.local collide on multicast DNS, and the losers rename themselves phaser-2, phaser-3, and so on, so a name you type can reach someone else’s kit or none at all.

The yes in Step 18 appears again because the laptop has never seen a kit named phaser-NN before. Running the installer again later is how a kit is updated: curl -fsSL https://raw.githubusercontent.com/livethisdream/phaser/main/install.sh | bash.

Read the log rather than systemctl is-active phaser-headless: a backend that fails at start-up is restarted every five seconds and reads active between failures. 192.168.2.1 is the Pluto’s own address on the USB link between it and the Pi.

If the pill never reads Connected, turn on Show Logs Tab under Plot Options and read the Logs tab: a message such as “Start blocked until backend is ready” says the backend is still starting, and the page will connect on its own once it is up.

The FFT tab refreshes once per sweep. Freeze stores up to three reference traces on the Rectangular and Polar tabs, and pressing and holding it clears them; it is unavailable on the FFT tab.

One meter is the shortest far-field distance for this array. Lesson 5’s far-field boundary, with the aperture \(D = Nd = 8 \times 14\ \text{mm} = 112\ \text{mm}\) and \(\lambda = 28.5\ \text{mm}\), is

so moving the source closer than about 0.9 m raises the signal but curves the arriving wavefront enough to distort the patterns of later labs.

One meter clears that boundary, so the angles and beamwidths measured there are the far-field values. It clears it by only 12%, though, which is marginal for comparing peak levels between two source positions, a difference of under a dB. On my kit, the drop with the source moved to 30° read only 0.2 to 0.4 dB at 1 m and 0.7 dB at 1.2 m, against a predicted 0.6 to 0.9 dB, so the L18 Lab takes its scan-loss levels at 1.2 m (4 ft) or more.

Each step needs the one before it. Characterizing the IF needs a tone at a known frequency, and calibrating needs the kit tuned to the source at an IF it can hear. On my kit, calibration first ran at the IF that ships in the configuration file, 2.2 GHz, and heard nothing: on/off contrast 0.93 dB. That kit receives cleanly only from 1.2 to 2.175 GHz, while a second kit works at 2.2 GHz; the usable IF differs by about 200 MHz between two kits built alike, which is why every kit measures its own. At the IF it measured, 1.69 GHz, the same calibration ran to completion with 32.7 dB of contrast. On a kit that has never been characterized, Calibrate Phaser runs the characterization first, or stops and says so.

The frequency matters downstream because the steering phase depends on it. Lesson 18 derives the progressive phase

so the same steering angle needs a different phase at a different frequency. For a 30° steer the ramp is 84.9° per element at 10.1 GHz and 89.9° at 10.7 GHz, and a ramp computed for the wrong frequency points the beam at the wrong angle.

All three results are stored in calibration.json on the Pi and survive a reboot, so a kit is calibrated once and stays calibrated until it is moved to a different source.

The host-key warning is ssh doing its job. A host key identifies the machine, not the name, so a re-flashed card, a fallback card, or a classmate’s kit that answers to a name this laptop has seen before presents a key the laptop does not recognize, and ssh refuses rather than risk talking to an impostor. ssh-keygen -R removes the old key from the laptop’s known_hosts file, and the next connection asks about the new one.

Lesson 2’s Friis equation sets the peak,

doubling the distance lowers it by \(20\log_{10}2 = 6.0\ \text{dB}\), and misaiming the HB100 lowers its \(G_t\). On the receive side, Lesson 20’s broadside directivity of the uniform array is

at 10.525 GHz (7.7 at the workshop’s 10.3 GHz), the \(G_r\) of the Friis equation.

The floor in the FFT is the noise in one frequency bin, so the peak-to-floor separation also depends on the FFT length: a longer FFT narrows each bin, lowers the floor, and raises the separation with no change at the antenna.

On a working kit the sweep’s floor sits about 23 dB below the uniform-taper peak. A normal FFT peak with a sweep that stands well short of that, or that peaks far from 0° with the source at boresight, points to the calibration rather than the signal: check that Phase Control reads zero, then put the source back at boresight and run Calibrate Phaser again. Lesson 18 derives why the sweep peaks at the source, and the L18 Lab measures it on a protractor arc.

With a checkout of the Phaser repository and Python, python phaser_headless.py --sim serves the same page at http://localhost:8080.