USB System design

The design architecture of USB is asymmetrical in its topology, consisting of a host, a multitude of downstream USB ports, and multiple peripheral devices connected in a tiered-star topology. Additional USB hubs may be included in the tiers, allowing branching into a tree structure with up to five tier levels. A USB host may implement multiple host controllers and each host controller may provide one or more USB ports. Up to 127 devices, including hub devices if present, may be connected to a single host controller.USB devices are linked in series through hubs. One hub—built into the host controller—is the root hub.

A physical USB device may consist of several logical sub-devices that are referred to as device functions. A single device may provide several functions, for example, a webcam(video device function) with a built-in microphone (audio device function). This kind of device is called a composite device. An alternative to this is compound device, in which the host assigns each logical device a distinctive address and all logical devices connect to a built-in hub that connects to the physical USB cable.

Diagram: inside a device are several endpoints, each of which is connected by a logical pipes to a host controller. Data in each pipe flows in one direction, although there are a mixture going to and from the host controller.

USB device communication is based on pipes (logical channels). A pipe is a connection from the host controller to a logical entity, found on a device, and named an endpoint. Because pipes correspond 1-to-1 to endpoints, the terms are sometimes used interchangeably. A USB device could have up to 32 endpoints (16 IN, 16 OUT), though it's rare to have so many. An endpoint is defined and numbered by the device during initialization (the period after physical connection called "enumeration") and so is relatively permanent, whereas a pipe may be opened and closed.
 

USB endpoints actually reside on the connected device: the channels to the host are referred to as pipes

There are two types of pipes: stream and message pipes. A message pipe is bi-directional and is used for control transfers. Message pipes are typically used for short, simple commands to the device, and a status response, used, for example, by the bus control pipe number 0. A stream pipe is a uni-directional pipe connected to a uni-directional endpoint that transfers data using an isochronous,interrupt, or bulk transfer:

• isochronous transfers: at some guaranteed data rate (often, but not necessarily, as fast as possible) but with possible data loss (e.g., realtime audio or video).
• interrupt transfers: devices that need guaranteed quick responses (bounded latency) (e.g., pointing devices and keyboards).
• bulk transfers: large sporadic transfers using all remaining available bandwidth, but with no guarantees on bandwidth or latency (e.g., file transfers).

An endpoint of a pipe is addressable with a tuple (device_address, endpoint_number) as specified in a TOKEN packet that the host sends when it wants to start a data transfer session. If the direction of the data transfer is from the host to the endpoint, an OUT packet (a specialization of a TOKEN packet) having the desired device address and endpoint number is sent by the host. If the direction of the data transfer is from the device to the host, the host sends an IN packet instead. If the destination endpoint is a uni-directional endpoint whose manufacturer's designated direction does not match the TOKEN packet (e.g., the manufacturer's designated direction is IN while the TOKEN packet is an OUT packet), the TOKEN packet is ignored. Otherwise, it is accepted and the data transaction can start. A bi-directional endpoint, on the other hand, accepts both IN and OUT packets.

Endpoints are grouped into interfaces and each interface is associated with a single device function. An exception to this is endpoint zero, which is used for device configuration and is not associated with any interface. A single device function composed of independently controlled interfaces is called a composite device. A composite device only has a single device address because the host only assigns a device address to a function.

Two USB standard A sockets on the front of a computer

When a USB device is first connected to a USB host, the USB device enumeration process is started. The enumeration starts by sending a reset signal to the USB device. The data rate of the USB device is determined during the reset signaling. After reset, the USB device's information is read by the host and the device is assigned a unique 7-bit address. If the device is supported by the host, the device drivers needed for communicating with the device are loaded and the device is set to a configured state. If the USB host is restarted, the enumeration process is repeated for all connected devices.

The host controller directs traffic flow to devices, so no USB device can transfer any data on the bus without an explicit request from the host controller. In USB 2.0, the host controller polls the bus for traffic, usually in a round-robin fashion. The throughput of each USB port is determined by the slower speed of either the USB port or the USB device connected to the port.

High-speed USB 2.0 hubs contain devices called transaction translators that convert between high-speed USB 2.0 buses and full and low speed buses. When a high-speed USB 2.0 hub is plugged into a high-speed USB host or hub, it operates in high-speed mode. The USB hub then uses either one transaction translator per hub to create a full/low-speed bus routed to all full and low speed devices on the hub, or uses one transaction translator per port to create an isolated full/low-speed bus per port on the hub.

Because there are two separate controllers in each USB 3.0 host, USB 3.0 devices transmit and receive at USB 3.0 data rates regardless of USB 2.0 or earlier devices connected to that host. Operating data rates for earlier devices are set in the legacy manner.