Voice packets resist interference

Bluetooth voice transmission uses synchronous connection-oriented (SCO) packets to provide full-duplex voice communication between two devices. The timing of voice and data transmission is organized around a slot framework. Each slot is 625 microseconds long, with six slots defining an SCO period. As voice is transmitted at 64 kbits/second, 240 bits are required to be transmitted during each SCO period. The high-quality voice (HV) packet types differ according to how much forward error correction (FEC) coding they use. The packet types range from HV1, employing one-third repeat coding and transmitting every second slot, to HV3, employing no coding and transmitting every sixth slot.

An additional access code and encoded header bring the packet up to 366 bits long, taking 366 microseconds to transmit using 1 Mbit/s Gaussian frequency shift-keying modulation. The receiver always replies in the next slot, giving three slot pairs in one SCO period.

Bluetooth uses frequency hopping, which entails changing to a new transmit frequency at every slot. In Europe and the United States, there are 79 frequencies, each 1 MHz apart. The pseudorandom-hop sequence, used during regular voice and data transmission, is designed to distribute the hop frequencies equally over the full 79-MHz ISM band during a short time interval.

Correct SCO reception is determined by checking the incoming packet's access code and header error check, which is a cyclic redundancy check in the header. Bluetooth Class 3 devices, currently the most common type of Bluetooth products, are intended to operate at low power and short range, transmitting at a maximum of 1 milliwatt.

Variable-length packets

But 802.11b transmission is very different from Bluetooth's. Packets are of variable length and can be more than 4,000 bytes long, equivalent to more than 50 Bluetooth slots. Transmission is asynchronous with the media-access control layer using a carrier-sense multiple-access with collision-avoidance technique similar to that used by Ethernet. The physical layer of 802.11b uses a combination of differential

binary phase shift keying, differential quaternary phase shift keying, Barker code spreading and complementary code keying in one of 11 possible overlapping channels, each with a fixed bandwidth of 22 MHz. The transmit power of 802.11b is also higher than Bluetooth's, with a maximum of 1 W in the United States.

Interference can occur when Bluetooth and 802.11b devices are in close proximity, and the Bluetooth frequency hops to within the 22-MHz 802.11b channel while the 802.11b device is transmitting. For example, if an 802.11b transmitter and a Bluetooth transmitter are equidistant from a Bluetooth receiver, the carrier-to-interference ratio, such as -2 dB, is considerably lower that the 18-dB signal-to-noise ratio usually required for a Bluetooth receiver. The Bluetooth FEC on the header or payload cannot protect the bits and the whole packet can be corrupted. The next packet of voice data will be transmitted at a new hop frequency, usually not in the same vicinity as the 802.11b transmitter.

The IEEE 802.15 working group on coexistence has proposed a new voice packet type, originated by Symbol Technologies' Steve Shellhammer, that is more immune to such interference. The Scort packet achieves more robust transmission by replacing bit-level redundancy of the FEC with packet-level redundancy. It repeats transmission of the same voice packet three times in one SCO period. Because the transmissions occur at three different hop frequencies, they are unlikely to all be within the 802.11b channel. It has no FEC on the payload as HV3 does, but transmits once every second slot like HV1.

Since the receiver component of the Scort algorithm has to react to events such as incoming packets and to track the state of successful reception and slot number, it is easily modeled and described as a finite state machine.

The receiver algorithm of Scort has to track two items of state: the slot-pair number and whether the packet has been successfully received. Here, these two items of state are modeled together in one hierarchical state machine. This method is particularly useful for visualizing the complete state of the receiver.

Good state

If the first packet that arrives is received correctly, the machine changes to the state "Slot_pair_1," substate "Good." It also latches and stores that packet. Movement between states is controlled by placing conditions on the transitions, which must be true to allow the state to change. In this case, the "Slot_OK" Boolean value is a result of testing the access code and header error check (HEC). If the packet was not received correctly due to a failed HEC or access code, the machine would transition to the state "Slot_pair_1", substate "Bad". At each of the next two slot-pair sample times, the state machine moves to the next slot-pair state, thus tracking the current slot.

It therefore knows when to reject a payload that has reached the third slot pair without receiving a correct packet. If the packet arrives at the "Good" substate of the last slot pair, the system accepts the payload of the good packet that it has stored.

In an interference scenario such as the one described, the Scort packet has a significantly lower frame error rate than other HV packets. However, just like HV1 packets, there is only enough capacity for one full-duplex voice link. Interoperability performance in unlicensed bands, such as the ISM for example, is difficult to predict. Simulation helps to accurately calculate behavior and understand how well devices will work together.

See related chart