Bluetooth radio designs employ a number of system architectures, from conventional intermediate-frequency (IF)-based systems with analog modulation to digital in-phase/quadrature (I/Q) modulator/demodulator configurations. Currently, various forms of modules are being used. Ultimately, circuit-level integration may be required for the lowest BOM. Regardless of how the Bluetooth design is configured, numerous issues must be addressed, including global regulatory requirements, Bluetooth certification, development of simple, high-yield manufacturing and test procedures, and flawless interoperability with designs from other vendors, some of which may perform at the limits of the Bluetooth specification. This article will examine different features of Bluetooth designs, implications for research-and-development (R&D) tests, and the tools that can make development easier. It will also describe how to performance RF measurements and what types of results should be expected.

Bluetooth devices operate in the industrial-scientific-medical (ISM) band from 2.402 to 2.480 GHz, usually on 79 pseudorandomly chosen channels, spaced 1 MHz apart. There are occasions, such as during the inquiry phase, when a reduced set of channels are used. Figure 1 shows how the spectrum usage changes during this period. Bluetooth devices communicate using a digital frequency-modulation (FM) technique known as 0.5BT Gaussian frequency-shift keying (GFSK). This means the carrier is shifted up nominally 157 kHz to represent a digital 1 value or down to represent a digital 0 value, at a rate of one million symbols (or bits) per second. The "0.5" confines the 3-dB bandwidth of the data filter to 500 kHz, thereby setting a limit to the RF spectrum occupied.

Bluetooth employs time-division-duplex (TDD) communication: the transmitter (Tx) and receiver (Rx) alternate transmissions in separate timeslots, one after the other. In addition, a frequency-hopping scheme, with up to as many 3200 hops/s during the inquiry phase, increases the reliability of a Bluetooth link in relatively crowded RF band. Optimum link performance is essential given that recent US Federal Communications Commission (FCC) rulings anticipate that band usage will almost certainly increase.

Figure 2 shows possible timings for sending and receiving a 366-µs DH1 packet, relative to the 625-µs timeslots. The lower traces indicate a settling-time interval. During this interval, the device must hop to the next channel frequency and the voltage-controlled oscillator (VCO) must settle in time for the packet data to be transmitted or received. The start of the packet is not directly related to the rising edge of the RF burst, as shown by the dotted lines representing possible alternative rising edges. Nor is the rising edge of the burst related to the beginning of the timeslot. After transmission of all the packet data, the design may ramp the power down immediately, or wait until near the end of the timeslot. The exact way the burst is sent may impact other Rx designs and the battery current used.

Bluetooth Receiver Layout
An example Bluetooth Rx layout (Fig. 3) employs one downconversion stage (the orange boxes are areas where parts are omitted or swapped in different designs) and a single local oscillator (LO). The output of the LO is frequency doubled and switched between the receive and transmit functions. The use of FSK allows simple direct modulation of the VCO. Baseband data is passed through a Gaussian filter, which is characterized by a constant time delay and no overshoot. Pulse shaping is applied only to the Tx. Either a sample-and-hold (S/H) circuit or a phase modulator can be employed to override attempts by the phase-locked loop (PLL) to strip off phase modulation within its bandwidth. Often the IF will be quite high, to limit the physical size of the filter components and to make sure that the IF is spaced far enough from the LO frequency for proper image rejection. Antenna switching is used when the transmit level would otherwise be high enough to overloading the Rx input. Occasionally separate transmit and receive antennas are used. The reference oscillator is usually included in most combined RF/Baseband modules.

An output amplifier is optional in Bluetooth systems, but if included it will be employed to boost the power required for Class 1 (+20-dBm) versions. The amplifier's level accuracy specification is not demanding, but some attention to power output is needed to avoid excessive power output and to minimize battery drain. Regardless of whether the design delivers +20 dBm or less, the Rx must be ready to provide received-signal-strength-indication (RSSI) information, so that devices from different power classes can operate. To equalize the link budget with a non-class 1 device, it may be necessary to employ an Rx that is more sensitive than required by the basic specification. Power ramping in a design such as this can be readily achieved by controlling the amplifier bias currents, but care needs to be taken with the modulation applied during ramping, otherwise unwanted spectral components can be generated.

Page Title

Bluetooth Spectrum Tests
Unlike other time-division-multiple-access (TDMA) systems such as Digital European Cordless Telecommunications (DECT) or Global System for Mobile Communications (GSM), Bluetooth spectrum tests are not gated to separate out individual spectrum components. The measurement interval must therefore be long enough to capture effects due to both ramping and modulation. This may not cause certification problems, but the availability of time-gated measurements is likely to be invaluable because of their ability to quickly identify defects. An even more powerful measurement is the spectrogram, which shows spectrum as a function of time. As shown in Fig. 4, some designs make use of the unspecified period before modulation begins, usually to prepare the Rx. In this example, neither a 1 nor a 0 is transmitted.

All the frequency measurements in the Bluetooth specification rely on short gate periods. This adds noise to the results, because the narrow time window gives a higher cutoff frequency to the measurement bandwidth, thus including a variety of noise mechanisms in the measurement. An allowance for this fact will need to be made in the design limits, beyond the static error contributed by the crystal reference.

Frequency Drift
Frequency Drift measurements provide a combination of short-term, adjacent data groups, and long-term, drift-across-the-burst results. Initially this was intended to address possible S/H PLL design weaknesses. For other designs, unwanted modulation components from around 4 kHz to 100 kHz, or noise, may be viewed graphically as ripple. Setting the payload to all 1's is one way of confirming that the power supply is properly decoupled.

In the Tx path, the VCO (Fig. 2) is directly modulated. To avoid the PLL stripping off modulation components inside its bandwidth, it is either opened during transmission, or phase-error correction is applied (two-point modulation). The S/H technique can be effective, but requires care to avoid frequency drift. The phase modulator must be calibrated to avoid a lack of flatness in the modulation response to different data patterns, unless a digital technique is employed to adjust the synthesizer division ratios. One of the modulation patterns used for certification tests appears in Fig. 5.

The Bluetooth RF specification checks the peak frequency deviation for two different patterns, 11110000 and 10101010. The output of the GMSK modulation filter reaches its maximum after 2.5 bits; the first pattern checks this. The cutoff point and shape of the GMSK filter are checked by the second pattern.

Ideally, the peak deviation of the 1010 pattern is 88 percent of the 11110000, although some designs transmit without 0.5BT Gaussian filtering applied and so will show higher ratios. The highest fundamental modulation frequency is 500 kHz, even though the bit rate is 1 Msymbol /s.

The "20-dB test" confirms that a modulated and pulsed Bluetooth signal fits within a 1-MHz-wide band. Because of amplitude pulsing in the test signal, the measurement is performed with the "peak-hold" function, which makes allowance for the waveform being off the exact center frequency, by making it a "frequency-width" test, rather than just a fixed mask. The effect is very similar if the signal is centered within the mask. "Bumps" can appear in the spectrum due to non-datawhitened zeros or ones in the header of the packet.

The Bluetooth specification calls for adjacent-channel measurements to be performed as a series of spot-frequency measurements. An un-gated sweep is a fast, easy way to check for problems.

Frequency doubling is commonly employed to prevent RF from coupling back to the VCO, thereby causing pulling of the center frequency. Sub-harmonics must be removed from the RF output path, particularly if there is a danger they will affect the performance of co-sited functions.

Figure 6 shows a signal from a design that exhibits no subharmonics, but with harmonics extending to 9 GHz. International regulatory specifications determine acceptable limits, rather than Bluetooth itself. Such a harmonic measurement can be performed with a standard spectrum analyzer. The 401-point sweep required 3.125 s. For investigative work, faster sweep times can be used, but they still require several seconds. If a long sweep time is chosen, newer spectrum analyzers, with deep, data-capture buffers, enable post-sweep zooming to specific points of interest.

Figure 7 shows that a number of designs have moved to I/Q mixing in both the Tx and Rx paths. This has the advantages of increasing the level of circuit integration and handling some functions with digital signal processing (DSP) rather than analog circuitry. Figure 7 depicts a hybrid approach. In some designs, image-rejection mixing is added at the front end. The high levels of silicon-level integration make this more affordable. A bandpass filter is common in the receive path.

Calibration of all these I/Q stages must be carefully accounted for, because their performance may drift. Published techniques from radar and cellular applications describe sequences and signals that can be used. Fortunately for most developers, these issues are taken care of by the chip designer. Direct application of I/Q modulation to the RF output can have a surprising effect on the signal. There may be no effect from misalignment of the modulators on the "frequency" component errors, since frequency is simply the rate of change of the phase. However, it may be difficult to discern errors in the spectrum. Errors in I/Q modulation mean that there is amplitude modulation present. This can be detected by using a power-versus-time display—or by using a vector analyzer to perform more detailed investigations.

Page Title

The I/Q modulator is often also used to shape the power ramp, again pointing to the potential value of gated measurements. As mentioned in the opening remarks, many forms of Bluetooth modules are found. In the receive chain, bit-error measurements will require some of the digital processing to be present before measurements are possible, or at least a level detector to act as a digital decision circuit. A zero-IF system, identified by looking for a DC block between the Rx mixer output and analog-to-digital-converter (ADC) input, may be implemented. Here, imperfections such as LO-RF feedback, which generates a DC component that varies with input frequency, must be carefully managed. Near-zero IF, often with the IF set at one-half the RF channel bandwidth, is more likely to be used. Then sideband suppression is an issue: a 0.1-dB gain error or 1-deg. phase error develops sidebands that are approximately 40 dB down.

By its nature, a vector analyzer can demodulate a wide range of signals. Though situations involving only directly applied frequency-shift keying (FSK) may not warrant the extra sophistication, the arguments are changing as more I/Q designs are used, or when other formats, such as Bluetooth 2, cellular radios, or wireless-local-area-network (WLAN) systems, are also being considered. Very powerful debugging techniques, such as RF data recording are now available by combining a conventional spectrum analyzer with external software.

Higher levels of integration are focusing attention on simulation tools. Not only do they enable different circuit topologies to be assessed quickly, but also the more advanced tools are able to inject a wide variety of valid and impaired signals into the Rx.

For Bluetooth, this is where some of the largest RF challenges may lie. With battery consumption in mind, the effects of restricted level-compression performance can be tested, along with phase noise, different path losses, signal impairments, and interference—including the effect of nearby Txs—as will be encountered when Bluetooth units are coupled to cellular telephones.

There are two aspects of more recent product developments that will be of significant benefit. The first is the integration of digital signal generation and vector signal-analysis blocks that provide for interchange between simulation and practical testing. Seamless links between software-analysis tools and physical test instruments allow prompt comparison of results when prototypes are delivered. The second feature is the provision of design guides that automate the set up of the tools. This provides the user with a significant boost toward using the design software for real circuit evaluations an alternative to programming in basic configuration information relating to a specific radio technology.

Front-end amplifier design and testing must focus on interference rather than the best possible noise figure or 1-dB-compression characteristics. Various published techniques are available to dynamically change gain through the Rx chain, thus optimizing the rejection of unwanted signals. Applying synchronous, pulsed amplitude modulation to the signal generator may be a worthwhile test of the burst-to-burst response of the automatic-gain-control (AGC) system, particularly if it is software controlled.

Testing the receiver of a complete Bluetooth module is done most conveniently using an Integrated Test Set. For bit-error-rate (BER) testing, a loopback signal path is created using a special test-mode command. Test packets received by the Device Under Test (DUT), are re-transmitted on the following burst. Figure 8 shows a typical user interface for a such a test. This, and the other test mode commands may be sent "over the air," but the DUT should only respond to them when it has been locally set to do so. In general use, a Bluetooth device will not respond to these commands. It may, therefore, be convenient to test in "normal" mode, when a simple Packet Error Rate test is run. Although not part of the Bluetooth standard, it provides a simple indication of a good or bad device.

Loopback Test
There are several other reasons why a loopback test may not always be appropriate. The module may simply not include all the baseband control needed to establish a Bluetooth connection. Making a loopback method requires the bursts to be transmitted twice, so the test may take longer. Finally, the designer may wish to run an impairment test that is outside the scope of those contained in the Bluetooth specification. For these situations, Fig. 9 shows measurement paths for testing a Rx in isolation, which provides maximum flexibility for the designer.

As previously noted, a common element in all Bluetooth designs is the use of a single LO. This approach, however, requires that the LO can slew across the full tuning range in less than 300 µs. This must also occur when the device is operating in Bluetooth test mode. During the transmit period, a frequency may be chosen which is at the opposite end of the ISM band to the receive test frequency, or some other arbitrary point. The LO must make the transition back to the Rx frequency each time.

Since every burst can be used for data transmission, a continuous sequence can be employed for BER testing, which may eliminate a need to perform a frequency-hopped BER test. Unless the link signaling is available, however, an operator must arrange for simultaneous control of the test signal generator and the unit under test. Once the bits have been converted to a digital format, BER testing is feasible. There are a number of ways this can be done. A summary of techniques for BER testing is listed in the table.

As has been shown here, standard Bluetooth measurements such as a test of Bluetooth modulation characteristics and frequency drift verify the quality of normal, but when there are failures the cause may not at all obvious. This is especially true for I/Q designs. In addition to the standard measurements, there are merits to performing non-specified measurements, such as "settling-time" and spectrogram measurements. In any case, to produce a robust Bluetooth design as quickly as possible, a radio designer should have access to a comprehensive suite of simulation and measurement tools.