Feature

The evolution of Bluetooth systems has been fast and furious. Just a few short years ago, Bluetooth products were simply drawings on engineering white boards and projects housed in R&D labs. Now, after receiving much hype, Bluetooth products are beginning to creep into the market.

As Bluetooth moves from the white board to reality, a great deal of change has occurred in the development process. Designers are no longer worrying about simply making Bluetooth technology work. They are focusing their attention on developing cost-effective designs that will survive in the highly competitive consumer electronics market.

But to make the leap from the R&D world to the consumer market, test and measurement becomes an integral part of the Bluetooth development process. In fact, proper testing techniques can be the difference between the success or failure of a Bluetooth solution.

During development of the Bluetooth specification, the Bluetooth Special Interest Group (SIG) developed a set of standard tests that designers can employ to evaluate the performance of a Bluetooth design. Getting a grip on these tests is key during the development of a Bluetooth product.

Bluetooth Operation

Communication between two Bluetooth devices is time division duplexed (TDD), meaning that the transmitter and receiver alternate their transmissions in separate timeslots, one after the other. In addition, a very fast frequency-hopping scheme (up to 1,600 hops/s) is employed to aid the reliability of the link in what is likely to be a crowded band.

Figure 1 highlights a Bluetooth architecture with time-to-market as the driver. Since Bluetooth employs a TDD scheme, most system designs employ a single local oscillator (LO) that is frequency doubled and switched between receiver and transmitter. The use of FSK supports simple direct modulation of the voltage-controlled oscillator (VCO). Baseband data is passed through a Gaussian filter, which has a constant time delay and no overshoot. Pulse shaping is only applied to the transmitter.

In this Bluetooth design, a sample-and-hold (S/H) circuit or phase modulator can be used to override the phase-locked loop's (PLL's) attempts to strip off phase modulation (PM) inside its control bandwidth. Antenna switching can also be used when the level would otherwise be high enough to overload the receiver input.

The receiver layout in Figure 1 is a single downconversion approach. Often the intermediate frequency (IF) will be quite high in order to keep the physical size of filter components down and provide a good percentage of spacing from the LO for easier image rejection.

Measuring Bluetooth RF designs

Frequency measurements on Bluetooth signals need to identify the start of the data packet within the burst. This is done by locating the preamble. The preamble, a fixed pattern of 1s and 0s at the start of the packet, is used as a reference when locating other parts of the packet, most significantly the payload. The signal during the preamble and payload sections of the packet is then measured to calculate the modulation metrics.

Normally, whitening (or randomization) is applied during signal transmission to reduce discrete spectral components. This effectively scrambles the payload data but not the access code or header of the packet. To make measurements such as modulation characteristics or carrier frequency drift, where the payload must contain fixed patterns, it is essential that whitening can be switched off.

Testing a Bluetooth system with random frequency hopping adds an extra degree of complexity to the analysis of the signal. Although it is important to test the hopping functionality of the device, it is not necessary to measure all aspects of the signal in this mode. In fact, it is better to isolate different effects seen on the signal by eliminating multiple causes.

For example, the modulation characteristics test, which is used to verify the modulator performance and premodulation filtering, may be affected by the channel number, but is unlikely to be affected by frequency hopping. Bluetooth devices must therefore support a test mode that provides, among other things, the ability to disable frequency hopping.

Carrier frequency

ICFT is one of the more important measurement techniques defined by the Bluetooth SIG. The ICFT test is a specified method of verifying frequency error in the VCO.

The ICFT measurement is concerned only with the frequency at the start of the framed packet data, as opposed to just after the rising edge of the RF burst. Figure 2a shows the 1010 preamble pattern circled, with a vertical line positioned to denote the center of bit p0. The start of bit p0 is a zero crossing, and is defined as the point in time 68 bit periods before the zero crossing that immediately precedes the access code trailer. The ICFT measurement method requires the frequency deviation values to be integrated over the packet's four preamble bits, and the result shall be taken to be the carrier frequency (f0). The error is then the difference between the carrier frequency f0 and the nominal carrier frequency.

The Bluetooth ICFT test specification does not define exactly where integration over a packet's four preamble bits should start and stop, but it is important that the measurement works for the 1010 and 0101 pattern preambles (see Figure 2b). If engineers first start by integrating over a 4-microseconds period, to get a nominal zero integral the engineers need to make sure that they include equal positive and negative areas.

Problems can occur when designers attempt to integrate 4 microseconds of the 0101 preamble, because designers cannot be certain of the carrier's state immediately prior to the preamble. For this reason, it has been shown that completing the calculation under these circumstances can result in an error up to 8% of the 75-kHz specification limit.

A solution to the problem created by integrating 4 microseconds of the preamble is to use a 3-microseconds integration length, offset by a half bit from the original p0 center position. This method ensures that we can integrate over equal positive and negative areas, regardless of the preamble type, and therefore avoids having an offset in the result. It is important to note that some additional noise is introduced when using the 3-microseconds integration length because the sampling time is reduced.

Measuring drift

Drift measurements provide a combination of short-term (10 bit, adjacent data groups) and long-term (drift across the burst) results. In Bluetooth architecture employing a sample-and-hold PLL design, frequency drift errors may be readily apparent. For other designs, unwanted modulation components from approximately 4 to 100 kHz and noise may be seen as ripple on the graphical result.

As defined by the Bluetooth SIG, frequency drift measurements require the payload data to consist of a repeating 4-bit 1010 sequence. The ICFT from the preamble is calculated as described above and stored as the reference frequency. Each successive whole 10-bit section of the payload is then analyzed using the same integration method as was used for the preamble. The analyzed section is then compared to the reference frequency.

Maxing out

If a linear frequency ramp is applied across the burst, there are two reasons why expected results will not match the transmit measurement exactly. First, the payload data length may not be an exact multiple of 10. Therefore the measurement will not get right to the end of the burst. Second, integrating over 10 bits actually yields the frequency in the middle of the 10-bit slot.

Therefore, even if the payload length were divisible by 10, designers would lose 5 bits off the end of the payload, and for the same reason, 2 bits off the start of the preamble. Hence, for a DH1 packet of length 366 microseconds, engineers would lose 7 microseconds of length off the measurement. Therefore, at best, a designer could expect to see approximately 98% of the applied impairment.

Evaluating deviation

The Bluetooth specification checks the peak frequency deviation for two different patterns: 11110000 and 10101010. The output of the GFSK modulation filter reaches its maximum after approximately 2.5 bits. This is checked by the 11110000 pattern. The cutoff point and shape of the GFSK filter are checked by the 10101010 pattern. Ideally, the peak deviation of the 10101010 pattern is 88% of the 11110000. As will become apparent, the highest fundamental frequency of modulation is 500 kHz, even though the bit rate is 1 Msymbol/s.

The light gray trace in Figure 3 indicates what happens to the 11110000 waveform if an in-phase & quadrature (I&Q) gain imbalance exists in the modulator. The modulator used here was programmed with a calibrated impairment of 2-dB I&Q gain. The effect can be understood by considering the trajectory of the sum vector in the I&Q plane. Instead of the usual circle, an oval is formed. As the sine and cosine I&Q components move up and down their respective axes, the angle of the sum vector no longer directly matches the I&Q values. The angular velocity is therefore changing and hence variations in FM are introduced.

The shape seen on the frequency deviation trace depends on the type of I&Q impairment. For the 11110000 pattern, the peak deviation is in-creased by approximately 25%, while the deviation of the 10101010 pattern is reduced by about 12%. This results in an overall deterioration in the ratio of the two deviations to the point of failing the Bluetooth specification.

If the same signal is now measured using frequency drift or ICFT measurements, the impairment goes undetected due to the symmetrical nature of the impaired waveform and the type of averaging used.

A generic approach

Fortunately, a new deviation ratio measurement has been under development. This new approach makes use of the fact that when fully charged, a 0.5-BT Gaussian filter will provide a maximum deviation after approximately 2.5 bits. Therefore, as a designer scans the frequency deviations representing each bit from left to right, any series of three 1s or three 0s can be used to contribute to the 11110000 deviation value. Similarly, the middle bit of each detected 101 or 010 sequence can be used to contribute to the 10101010 deviation value.

The accuracy of the new modulation characteristics deviation method has been shown to be within 2% of the certification method described above. This allows some flexibility in the choice of test pattern to maximize the required occurrences, and would even cope with whitened payloads.

Integrated bluetooth designs

Figure 4 shows something of a hybrid approach that employs I&Q mixing. In this design, image rejection mixing is added at the front end. Engineers should note that the calibration of all I&Q stages needs to be carefully accounted for. Published techniques from radar and cellular applications describe possible sequences and signals.

Direct application of I&Q modulation on the RF output can have a surprising effect on the signal. It may be difficult to discern errors in the spectrum. Errors in I&Q modulation can give AM. This situation can be viewed using a power-versus-time display, or by using a vector analyzer for more-detailed investigations. The I&Q modulator may also be used to shape the power ramp, pointing to the potential value of gated measurements.

Let's take a look at measurements performed on a vector analyzer. By its nature, a vector analyzer can demodulate a very wide range of signals. While situations involving only (directly applied) FSK may not warrant the extra sophistication, the argument changes where I&Q designs are in progress, or when other formats, such as cellular or wireless LAN (WLAN) systems, are being considered. To understand the device's behavior, it becomes more important to be able to analyze it from multiple directions.

Figure 5 gives an example of the same data being viewed in four different ways. The deviation view gives a fast visual confirmation of the correct modulation pattern. The eye diagram and FSK error demonstrate modulation quality. The demodulated data view enables the user to check for the presence of preambles, headers, sync words, and payload data.

Testing with non-hopping test signals

During the transmit period, a frequency may be chosen that is at the opposite end of the ISM band to the receive test frequency, or some other arbitrary point. The VCO has to make the transition back to the receiver frequency each time.

Every burst can be used for data transmission, so a continuous sequence can be used. The net effect is a less obvious need for performing a frequency-hopped bit-error rate (BER) test, where the source has to hop. While this method can be used, the designer needs to arrange simultaneous control of the signal generator and the device under test (DUT) until the link signaling is available.

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 is shown in Table 1.

Table 1: Methods Used for Receiver Bit-Error Measurement
Data recovery point	Description
IF	Use eye diagram
Demodulator output	Gate a raw PN sequence output and send for BER measurement.
Baseband output	Recover clock, decoded payload data, Send for BER measurement.
Loopback	A complete unit requires use of Bluetooth test mde. Baseband and link processing must be included. Some designs allow it to be done with custom settings.

There is no question that Bluetooth products are primed to take the communication and consumer markets by storm. But to make these products a reality, designers must develop systems that achieve the performance, battery-life, and price demands of modern end users. Through proper testing, developers of Bluetooth systems can detect problems earlier in the design process and develop products that will succeed in the communication market.

About the Author

Alistair Mill is a senior development engineer specializing in video and RF communications with Agilent Technologies. He joined Agilent as a software engineer after graduating in 1992 from Glasgow University with a BSc (Honors) in computing science with electronics. He can be reached at alistair_mill@agilent.com

Figure 1: Block diagram of a Bluetooth radio employing a direct frequency modulated VCO architecture.
Figure 2: a.) ICFT tests verifying frequency error in the VCO. b.) ICFT measurements must work...
Figure 3: Diagram of the impact of imbalance on a modulator employed in a Bluetooth design.
Figure 4: Bluetooth design employing an I&Q modulator and a digital demodulator.
Figure 5: Multiple views of FSK.

Table 1:Methods used for receiver bit-error measurement.

Return to the Table of Contents