Field notes from electronica 2026, Munich.
Disclosure: I work for UNI-T (UNI-TREND Technology EU GmbH). This demo ran on our own stand. The readings below are what the instruments showed during the show, not datasheet claims.
Key facts
- Instrument: UNI-T MSO6204HDP (2 GHz, 10 GSa/s, 12-bit) with a USB 2.0 signal-quality fixture.
- Stimulus: device put into
TEST_PACKETmode with the xHCI HS Electrical Test Tool. - Read from the screen: TIE 107.6 ps; total jitter at BER 1e-12: 228.3 ps; 231.7 million UI in the eye.
- Calculated: 1 UI at 480 Mb/s = 2.083 ns, so TJ is about 11 % of a UI.
- Not quoted: the RJ/DJ split, because it does not pass a dual-Dirac consistency check.
The first of seven demo stations on the UNI-T stand was labelled "USB Signal Integrity: Eye, Jitter, Compliance". The setup is small: a USB 2.0 signal-quality test fixture, a laptop and an MSO6204HDP oscilloscope (2 GHz, 10 GSa/s, 12-bit). The screen showed a full jitter analysis on a live high-speed USB link: jitter table, TIE histogram, TIE trend, jitter spectrum, bathtub curve and eye diagram, all at once. This article walks through how the signal is created, what each pane answers, and which numbers I can and cannot vouch for from my photo.
Step 1: make the device send a known pattern
A USB 2.0 high-speed link carries no repeatable pattern in normal traffic, so electrical testing starts by putting the device under test into a test mode. On the laptop next to the scope, the xHCI HS Electrical Test Tool listed the devices on the bus, one was selected, and the command TEST_PACKET was executed ("Operation Successful"). In this mode the device transmits the standard test packet continuously: a fixed sequence with long runs and fast transitions, which exercises both inter-symbol interference and edge timing. Every compliance measurement on the scope relies on this step. If the device is not really in test mode, the eye and jitter results describe whatever happened to be on the bus.
Step 2: read the six panes in the right order
Eye diagram. The eye panel reported a recovered bit rate of 481.3 Mbps (nominal 480 Mb/s) and a UI count of 231,699,769: more than 230 million unit intervals folded into one picture. The vertical scale ran to about ±400 mV, which is the region of the 400 mV nominal high-speed differential swing. The eye answers one question: is there margin? It cannot tell you where any lost margin went.
TIE trend. Time interval error, each edge's deviation from its ideal position, plotted against time. The trace showed a dense band with a slow wander. TIE is the raw material for everything below.
TIE histogram. The distribution of TIE values, here a peaked shape centred on zero. Its tails are what the software fits to separate random from deterministic jitter.
Jitter spectrum. The FFT of the TIE record. Discrete lines are periodic jitter, and their frequencies are clues. The two cursor labels on this pane read 24.066 MHz and 120.32 MHz. A 24 MHz crystal is a common USB reference, and 120 MHz is a quarter of the line rate, so a line near either frequency is a natural place to start looking for a coupling path. That is my reading of the labels, not a diagnosis of this board.
Bathtub curve. Bit-error ratio against sampling position, down to 10⁻¹⁴ on the axis. The horizontal opening at the target BER is the timing margin you actually have.
Jitter parameter table. TIE, TJ at BER 10⁻¹², RJ, DJ, PJ, DDJ, DCD, ISI, cycle-to-cycle and cycle jitter, each with a note.
Which numbers I will quote, and which I will not
From the photo I can read a TIE of 107.6 ps and a total jitter at 10⁻¹² of 228.3 ps. I am not quoting the RJ/DJ split. As photographed, the decomposition cells do not add up the way a standard dual-Dirac model would (TJ = DJ + 2·Q(BER)·RJ, with 2·Q ≈ 14.07 at 10⁻¹²). That can come from the decomposition model, the display or simply my reading of a tilted screen. Either way, the honest course is to confirm the setup before publishing a breakdown. This is the same rule I apply to every booth: a number that does not survive a sanity check stays out of the article until it does.
For scale, one USB 2.0 high-speed unit interval is 1/480 MHz ≈ 2.083 ns. A total jitter of 228.3 ps is therefore about 11 % of a UI.
Why the setup details matter
- Fixture and termination. The signal-quality fixture sits between host and device, so the scope sees the link at a defined reference point. Moving the probing point changes the eye.
- Vertical resolution. Edge times come from threshold crossings. Vertical noise on a sloped edge becomes timing noise (σt ≈ σv / slew rate), so a 12-bit acquisition with the trace filling the screen keeps the instrument's own contribution to random jitter low.
- Record length. The lowest periodic-jitter frequency the spectrum can resolve is set by how long the TIE record is. A short capture hides slow wander.
Checklist for a USB 2.0 high-speed jitter measurement
- Confirm that the device is in TEST_PACKET mode before trusting any eye.
- Probe at the fixture's reference point; keep the differential connection symmetric.
- Fill the vertical range; avoid traces that occupy two divisions.
- Read the jitter spectrum for discrete lines before looking at TJ.
- State BER and decomposition method with every TJ number, and check that TJ, RJ and DJ are mutually consistent.
Takeaway
A USB 2.0 high-speed jitter test is a chain: host tool, test mode, fixture, acquisition, decomposition. The screen at this station showed the whole chain at once, and that is its value for teaching. The eye says whether there is margin; TIE, spectrum and bathtub say where it went. Treat the decomposition table with the same scepticism you would apply to anyone else's instrument: check that the numbers agree with each other before you act on them.
More on this topic: Signal integrity and high-speed digital: jitter, eye diagrams, BER.
Notes: Observations are from the stand during the show on 10 November 2026. Values are as read from my photos. The RJ/DJ decomposition is deliberately not quoted until confirmed. The UI calculation is mine.
