How USB actually works
Plug in something the computer has never seen and a few hundred milliseconds later it works. In that gap the device is interrogated, assigned an address, asked to describe itself in detail, matched to a driver and switched on — a conversation in which the device is never once allowed to speak first.
The rule that shapes everything: only the host may start
USB's defining design decision is that devices have no right to speak. The host asks; the device answers. Nothing on the bus may transmit unless it has just been addressed, and nothing may interrupt.
This sounds restrictive and it is the reason USB works at all. With one controller deciding who talks and when, there are no collisions to detect, no arbitration to negotiate and no need for devices to be clever. A USB mouse can be built around a chip costing pennies because it never has to make a decision about the bus.
Interrupts that are not interrupts
A mouse appears to interrupt the computer when you move it. It does not — it cannot. What actually happens is that the host has agreed to ask the mouse 'anything to report?' at a fixed interval, often every millisecond. Most of the time the mouse says no. When it says yes, the movement is delivered.
This is called an interrupt transfer, which is a thoroughly misleading name — it is polling with a guaranteed maximum interval. It is also why a gaming mouse advertising a '1000 Hz polling rate' is describing exactly this: the host asking a thousand times a second instead of a hundred and twenty five.
Four kinds of traffic
| Transfer type | Used for | Guarantees |
|---|---|---|
| Control | Setup, configuration, commands | Always gets through; small |
| Interrupt | Mice, keyboards, sensors | Checked at a guaranteed interval |
| Bulk | Drives, printers, scanners | Correct eventually; no timing promise |
| Isochronous | Webcams, audio interfaces | On time, or dropped — never retried |
The last one is the interesting exception. Isochronous transfers have no error correction at all, deliberately. A late video frame is worse than a missing one, so the bus guarantees bandwidth and timing instead of accuracy. A corrupted audio sample produces a click; a retransmitted one would produce a stutter.
The takeaway One host asks; everything else answers. That single rule removes collisions, arbitration and complexity from every device on the bus, at the cost of devices never being able to talk to each other.
What happens in the moment you plug it in
Between the plug seating and the device working, a fixed and rather formal conversation takes place. It is called enumeration, it is the same on every operating system, and understanding it explains most of what goes wrong.
The port has two data lines held at a known voltage by resistors on the host side. A device announces its existence purely by pulling one of them high — there is no message, just a change in voltage that the host notices. Which line it pulls tells the host the device's speed before a single byte has been exchanged.
- Attach detected. The device pulls a data line high. The host sees the change and knows something is there, and roughly how fast it is.
- Reset. The host holds both lines low briefly, putting the device into a known state. It now has no address and responds only to address zero.
- First descriptor read. The host asks address zero for the start of its device descriptor, mainly to learn the maximum packet size it can use.
- Address assigned. The host allocates a number from 1 to 127 and tells the device to use it. Everything afterwards is addressed to that number.
- Full descriptors read. The host asks for everything: vendor and product IDs, class, interfaces, endpoints, power requirements, text strings.
- Driver matched. The operating system looks at what came back and chooses a driver — by class if it is a standard kind of device, by vendor and product ID if it needs a specific one.
- Configuration set. The host tells the device which of its configurations to adopt. Only now may the device draw its full current and begin normal operation.
The descriptors are the whole story
A device's descriptors are a small tree of structured data that describes everything about it. This is why USB devices are plug and play in a way that earlier buses never were: the device explains itself, and the host does not need prior knowledge.
What the vendor and product IDs really are
Every device reports a 16-bit vendor ID and a 16-bit product ID. Vendor IDs are allocated by the USB Implementers Forum for a substantial fee, which is why cheap no-name hardware frequently ships with a borrowed or invented one — and why two unrelated gadgets can occasionally confuse each other's drivers.
When Windows says 'USB device not recognised', this is where it failed: either the descriptors could not be read at all — usually a cable or power problem — or they were read and nothing matched. The two look identical to a user and have completely different fixes.
The takeaway A device announces itself with a voltage, is reset, given a number, and then made to describe itself in full. Plug and play is nothing more than a device that can explain what it is.
Why most things work without installing anything
Plug in a mouse from any manufacturer, on any operating system, made in any of the past twenty-five years, and it works immediately. That is not luck — it is a small number of standard classes that almost everything conforms to.
Rather than every manufacturer writing a driver for every platform, USB defines device classes: published specifications for how a category of device behaves. A device that declares itself a member of a class is promising to follow that specification, and the operating system already has a driver for it.
| Class | Covers | Notes |
|---|---|---|
| HID | Keyboards, mice, gamepads, dials | Self-describing report format — extremely flexible |
| Mass storage | Flash drives, external disks, card readers | Effectively a SCSI disk over USB |
| Audio | Headsets, interfaces, speakers | Class 2 gives high-resolution audio driverless |
| Video | Webcams, capture devices | UVC — why any webcam works anywhere |
| CDC | Serial adapters, some modems | Emulates a classic serial port |
| Printer | Printers | Transport only; the page language is separate |
| Hub | Hubs | Standardised so hubs never need drivers |
HID is cleverer than it looks
The Human Interface Device class does something unusual. Rather than defining a fixed format for a mouse and another for a keyboard, it defines a language in which a device describes its own controls. A device sends a report descriptor saying, in effect, 'I will send you four bytes; the first two are X and Y movement in this range, the next is a set of button flags, the last is a scroll wheel'.
The operating system parses that and now understands a device nobody anticipated. It is why a racing wheel, a foot pedal, a flight yoke and a 3D mouse all work without drivers despite having nothing in common with a mouse — and why the 'no driver needed' promise has survived twenty-five years of new input devices.
When a manufacturer driver is still needed
Beyond the class
The device works as a standard class member, and the manufacturer's software adds macros, RGB lighting or configuration the class does not cover.
Custom protocols
High-end audio interfaces often bypass the class driver for lower latency, which is what ASIO drivers are doing.
No matching class
Specialised instruments and industrial hardware that no class describes.
Cheap serial chips
Some USB-to-serial chips deviate from CDC to save a few pence, which is why they need a specific driver and why counterfeits of them cause such trouble.
The takeaway Standard classes mean the operating system already has the driver. HID goes further and lets a device describe controls that did not exist when the standard was written.
How much power it may take, and how it asks
USB began as a data bus that could spare a little power for a mouse. It now charges laptops. The route from one to the other is a series of negotiations, and a device that does not negotiate correctly is the reason your phone sometimes charges slowly.
Original USB offered a strict budget: 100 milliamps at 5 volts on plugging in, and up to 500 mA — two and a half watts — once configured and granted it. A device exceeding its allowance could be cut off. That was ample for a mouse and hopeless for anything with a battery.
How Power Delivery negotiates
USB-C has a dedicated configuration channel — the CC line — used for exactly this. A charger advertises a menu of what it can supply: 5 V at 3 A, 9 V at 3 A, 15 V at 3 A, 20 V at 5 A. The device picks the entry it wants and requests it. The charger confirms and switches.
| Voltage | Common current | Power | Typically used by |
|---|---|---|---|
| 5 V | 3 A | 15 W | Phones, accessories, legacy |
| 9 V | 3 A | 27 W | Fast phone charging |
| 15 V | 3 A | 45 W | Tablets, small laptops |
| 20 V | 5 A | 100 W | Laptops — needs a marked cable |
| 28–48 V | 5 A | up to 240 W | EPR: large laptops, monitors |
If either end does not support Power Delivery, they fall back to 5 volts and whatever current the resistors on the CC line indicate. This is why a laptop charger can charge a phone slowly: the negotiation failed or was never attempted, so both ends defaulted to the safe minimum.
The cable is part of the negotiation
Above 3 amps the cable itself must identify its capability. Cables rated for 5 A contain a small chip — an e-marker — that the charger reads. Without it, the charger will not supply more than 3 A no matter what both ends could manage. This is a genuine safety measure: a thin cable carrying 5 A becomes a heating element.
It is also the single most common cause of disappointing charging. A 100 W charger, a 100 W laptop and an unmarked cable from a drawer gives you 60 W at best. The cable is doing exactly what it should.
Bus-powered devices and where they fail
An external drive that spins a platter needs more current at spin-up than at rest. Plugged into a port that cannot supply the surge, it will click, fail to mount, or mount and then disconnect under load. An unpowered hub makes this worse by sharing one port's budget among everything plugged into it — which is why the fix for a misbehaving drive is so often a powered hub.
The takeaway Power is granted, not taken. The device asks, the charger offers a menu, and the cable gets a vote — which is why the cable you use changes how fast things charge.
USB-C is a connector, not a capability
The oval connector is the most visible change USB has made, and the most misunderstood. Two identical ports on two laptops can differ enormously in what they do, because the shape tells you nothing about what is wired behind it.
USB-C solved several real problems at once. It is reversible, so there is no wrong way up. It is symmetrical, so the cable has no ends. It carries far more power. And it has enough pins to carry signals that are not USB at all.
Alternate modes: other protocols on the same pins
The connector has four high-speed lanes. An alternate mode is an agreement, negotiated over the CC line, to hand some or all of those lanes to a completely different protocol. The port stops being a USB port for the duration.
DisplayPort alt mode
The lanes carry a video signal directly. This is how a single cable drives a monitor, and why a USB-C to HDMI adapter is mostly just a converter, not a graphics device.
Thunderbolt
Carries PCI Express and DisplayPort at once — effectively an external extension of the machine's internal bus. External GPUs and fast storage depend on it.
HDMI alt mode
Defined and then almost entirely unused; adapters won instead.
Analogue audio
Some phones route plain analogue audio over the connector, which is why certain USB-C headphones only work on certain phones.
Why this causes so much confusion
Every previous USB connector told you something. A rectangular Type-A port was USB and only USB. The shape was a guarantee. USB-C deliberately broke that link to avoid a drawer full of different cables, and the cost is that the port no longer tells you what it does.
It is worth checking the specification of a machine rather than counting ports. A laptop with four USB-C ports may support video on two of them, high-power charging on one, and Thunderbolt on none.
The takeaway USB-C is a shape that can carry USB, DisplayPort, Thunderbolt and power in any combination the manufacturer chose to wire up. Identical ports are routinely not equivalent.
Making sense of the version numbers
USB's naming is genuinely, notoriously bad — repeatedly renamed retroactively, so the same hardware has had three official names. Here is the translation table, which is more useful than any explanation of the reasoning.
| Speed | Original name | Renamed to | And again | Marketing name |
|---|---|---|---|---|
| 480 Mbps | USB 2.0 | — | — | Hi-Speed |
| 5 Gbps | USB 3.0 | USB 3.1 Gen 1 | USB 3.2 Gen 1 | SuperSpeed |
| 10 Gbps | USB 3.1 | USB 3.1 Gen 2 | USB 3.2 Gen 2 | SuperSpeed 10Gbps |
| 20 Gbps | USB 3.2 | USB 3.2 Gen 2×2 | — | SuperSpeed 20Gbps |
| 40 Gbps | USB4 | USB4 Gen 3×2 | — | USB4 40Gbps |
| 80 Gbps | USB4 v2 | — | — | USB4 80Gbps |
Read across any row and you are looking at one piece of hardware with several names. A drive sold in 2012 as 'USB 3.0' is identical to one sold in 2020 as 'USB 3.2 Gen 1', and a buyer could reasonably conclude the second is three revisions newer. It is not.
The current advice
The standards body has now accepted the problem and asks manufacturers to label products by speed and power — 'USB 40Gbps', 'USB 240W' — rather than by version number. This is a substantial improvement and adoption is gradual.
- Ignore version numbers on packaging; they are unreliable and frequently retroactive.
- Look for the speed in gigabits per second, which is unambiguous.
- Look for the wattage separately — a fast data cable may be a poor charging cable and the reverse.
- For video, check explicitly for DisplayPort alt mode; it is not implied by any speed.
- For Thunderbolt, look for the lightning bolt symbol on both the port and the cable.
Cables are the real trap
A USB-C cable can legitimately be any of: charge-only with no data lines at all; USB 2.0 data at 480 Mbps with 240 W charging; 40 Gbps data with only 60 W charging; or full capability on both. All four look identical and none of them is defective.
The cable supplied with a phone is usually USB 2.0 — entirely adequate for charging and hopeless for an external SSD. This is the most common reason a fast drive performs badly: it is running at 480 Mbps through a charging cable, which is roughly forty times slower than it should be.
The takeaway Six steps, a bus where nothing may speak first, and power negotiated over a dedicated wire — and the thing most likely to disappoint you is a cable that looks exactly like the right one.