Analog Signal Generators Updated

The use of a word evolves over time, creating the possibility for confusion. Analog is a good example. In some industries, analog refers both to the circuit technique used to generate signals and to the signals themselves. This is the traditional use of the term for a signal generator that basically is an oscillator.

However, this is not the only way the word is used. When we asked industry experts to discuss their company’s analog signal generators, virtually all replied that they made analog generators. True, these instruments have analog outputs, but almost all are synthesized.

Technically, there are good reasons to build synthesized signal generators, and because of modern semiconductor technology, complex designs have become economically feasible. Nevertheless, there still exist traditional analog generators, although they tend to serve either very lax or equally stringent needs. At the low-cost end, designers will always have a need for a simple CW signal source. In time, however, semiconductor integration will make it cost-effective to replace all but the most basic analog generators with digital designs.

At the high end, analog generators retain several advantages for specific applications. Regardless of the exact technique used in a synthesizer, signal generation involves counters, dividers, and digital control logic. All these elements switch on and off at high rates, creating spurious signals and noise. In contrast, a true analog oscillator may not have six-digit frequency accuracy and long-term stability, but neither does it have the same level of noise and distortion.

Bruce Hofer, chairman and co-founder of Audio Precision, commented, “The residual distortion performance of state-of-the-art analog signal generators significantly exceeds that of other techniques. The analog signal generator used in the 2700 Series Audio Analyzer has at least 10- to 30-dB better distortion performance than the best available digital signal generators.”

Mr. Hofer’s statement confirms that the 2700 analog generator uses analog techniques. It also shows how the term digital has become confused. In addition to the analog generator, the 2700 provides a synthesized generator, which Mr. Hofer called a digital signal generator. Many other industries consider a digital signal generator to be an instrument that produces digital signals at its outputs.

For a 22-kHz measurement bandwidth, the 2700’s analog generator total harmonic distortion plus noise (THD+N) for outputs between 20 Hz and 20 kHz is specified to be less than -110.5 dB. This compares with -103 dB for the digital generator specified under similar conditions. One way to put these numbers into perspective is to compare them to the 96-dB dynamic range of a perfect 16-b CD. The digital generator’s -103 dB corresponds to 17.1 b and the analog generator’s -110.5 dB to 18.35 b.

Digital Inside, Analog Outside
In contrast to audio testing, communications signal rates must be very accurately set. For example, the bit rate of 802.11ae 10-Gb/s Ethernet is specified as 10.3125 Gb/s with ±100-ppm maximum variation. This level of accuracy is routinely provided by crystal-controlled oscillators. A frequency synthesizer maintains crystal accuracy and stability while facilitating fine frequency selection.

Although several approaches are possible, each with distinct advantages, designs based on direct digital synthesis (DDS), a direct synthesis method, or a phase-locked loop (PLL), an indirect method, are used most often. Whether a DDS or PLL design is more appropriate depends on the application’s requirements. Synthesizer characteristics include frequency range, setting resolution, transient response, noise, and spurious content. These parameters are interrelated, so high-performance synthesizer designs can become complex and expensive.

National Instruments’ Ryan Verret, signal generators product manager, commented, “We use PLL, DDS, and time-to-digital conversion as well as digital interpolation in our signal generators. These technologies allow you to phase lock multiple generators to a shared reference clock to achieve synchronization with less than 20-ps skew. In addition, the reference clock could be sourced by an external, low phase noise oscillator, imparting the high performance of the oscillator to the digital generator.

“DDS clocking supports very fine frequency resolution,” he continued, “and interpolation moves reconstruction images to higher frequencies so that analog filters used to suppress the images do not compromise passband performance. Digital gain control and DDS generation also enable glitch-free amplitude changes and phase-continuous frequency sweeps.”

Phase Noise
Phase noise is one of the more important synthesizer parameters and is caused by the accumulated effect of many small perturbations to an oscillator’s instantaneous frequency. Mathematically, it can be expressed as the summation of a large number of small FM modulations. Timing jitter is the time-domain equivalent of phase noise in the frequency domain. However, although most digital systems successfully withstand a small amount of clock jitter, only minute amounts of phase noise can have very large effects in communications systems.

For example, the orthogonal frequency domain multiplex (OFDM) subcarriers in a WiMAX system are generated relative to a local oscillator (LO) frequency. If the LO has significant phase noise, then all the subcarriers also will. The resulting effects could be as bad as interference with other transmissions and reduced receiver sensitivity. Achieving less than 1° total phase noise is a reasonable goal for a synthesized LO. Some specifications set 5° as a maximum.

The addition of phase noise to a sinusoidal signal is not linear, but for low levels of phase modulation, it can be treated as though it were. With this assumption, the FM modulation index ? for each of the many modulating sine waves in the mathematical phase noise model is equivalent to the peak phase deviation. When ? is small, several simplifying narrowband approximations can be made.

As long as the total phase modulation is much less than a radian, the spectrum corresponding to each modulating frequency ƒ_m will consist of a pair of sidebands offset from the carrier by ±ƒ_m and with amplitude ?/2 relative to the carrier. Because of the sideband symmetry that follows from narrowband approximations, phase noise can be expressed as a single-sideband value. Phase noise density, L_PM in units of dBc/Hz, is the ratio of power in a 1-Hz bandwidth relative to the carrier power measured over a range of frequency offsets from the carrier.

Within any 1-Hz bandwidth, L_PM can be considered constant. From Figure 1, the phase noise characteristic of an Applied Radio Labs 450-MHz voltage-controlled oscillator (VCO), L_PM 10 kHz either side of the carrier is about -121 dBc/Hz. This means that at 450.01 MHz the power in a 1-Hz bandwidth measures -121 dB relative to the carrier power. You would measure the same power within a 1-Hz bandwidth at 449.99 MHz. If L_PM is constant over a wider bandwidth, the total power simply is the product of the bandwidth times L_PM in a 1-Hz bandwidth. More generally, the area under the phase noise curve must be integrated to find the total noise in a wider band.

Figure 1. 450-MHz VCO Phase Noise
Courtesy of Applied Radio Labs

Because the modulation index b is inversely proportional to the modulation frequency, the curve in Figure 1 slopes downward at a 20-dB/octave rate. This behavior is typical of oscillators with low L_PM.

Figure 2 shows the effect of phase noise on the VCO’s 450-MHz output. This plot was made using a spectrum analyzer with a 3-kHz resolution bandwidth (RBW) filter. The height of the skirts either side of the center frequency corresponds to L_PM integrated by the moving 3-kHz window as it was swept across the 80-kHz range.¹

Figure 2. Spectrum Analyzer Display of 450-MHz VCO Output
Courtesy of Applied Radio Labs

PLL-Based Synthesizers
The PLL in Figure 3 is an integer-N loop, called that because the ratio between the reference input and VCO output has an integer value. Assume that N in this example is 100. By definition, if the reference frequency is 100 kHz, the VCO output will be 100x larger or 10 MHz. The phase noise associated with an integer-N PLL includes a 20 log(N) contribution in decibels, which can become significant for large values of N. In this example, 40 dB of the total phase noise would be caused by the 100x frequency multiplication.

Figure 3. Integer-N PLL Block Diagram
Courtesy of Analog Devices

Loop bandwidth, set by the loop filter in conjunction with the other components, determines how fast the loop will lock to a sudden change in frequency. A narrow bandwidth locks more slowly but integrates less of the phase-frequency detector (PFD) noise and attenuates less of the VCO noise. A wide bandwidth has the opposite effect. The optimum bandwidth depends on the required performance and the amount of noise contributed by the different elements.

Once the VCO frequency is correct, the fastest PFD update rate to correct the VCO phase is the reference frequency, 100 kHz in the N = 100 example. This update rate generates a series of spurious frequencies at multiples of the reference frequency. Also, in an integer-N PLL, the output step size is constrained to be a multiple of the reference. This means that to have both a small step size and a high output frequency, N must be large.

Some integer-N synthesizer designs offset the VCO output by mixing it with a separate frequency before completing the feedback loop to the PFD. Mixing creates a lower feedback frequency to directly drive the PFD or at least reduce the required value of N if division still is needed. Of course, this scheme adds one or more stable and low-noise frequency sources and mixers, increasing cost and complexity.

In contrast, a fractional-N PLL achieves a lower value of N and lower L_PM by alternating between two different modulus dividers. For example, N = 19.1 results from dividing by 19 for 90% of the time and by 20 for 10% of the time. The advantage is that the same step size can be achieved by using a reference frequency 10x higher than if an integer N = 191 divider were used. To further increase flexibility, the reference input may be followed by one or more dividers ahead of the PFD, making a very large range of frequency ratios available.

Unfortunately, a regularly repeating alternation between divisors such as 19 and 20 introduces spurious frequencies. The PFD cannot lock to a single frequency but instead is locking to the average rate from the two N counters. Nevertheless, in many applications, reduced L_PM coincident with a lower value of N can make a fractional-N PLL design a good choice. This is particularly true if the spurious frequencies can be sufficiently filtered without significantly affecting other loop characteristics.²

Much recent PLL development has been based on the fractional-N architecture but altered to reduce or eliminate spurious generation. A popular method dithers the value of N using the sigma-delta technique. The sigma-delta process provides noise shaping, so a low-pass filter can eliminate much of the transient energy that has been shifted to high frequencies. The desired result is a uniform (white noise) increase to the noise floor rather than the addition of distinct spurious components.

In another approach, phase interpolation has been used to avoid or reduce spurious generation. In this technique, a small analog correction signal is added to the output of the PFD so the VCO will more closely track the desired output frequency. Using the N = 19.1 example, when dividing by 19, a small correction would be added to the PFD; when dividing by 20, a small correction would be subtracted. Because the correction signal must be closely controlled, modern DACs with small glitch energy and good linearity make this approach more feasible than in the past.

Yet another way to minimize noise while improving switching speed is used by the Giga-tronics Panther 2500 Series Frequency Synthesizers over a 100-kHz to 40-GHz range. Rather than a fixed reference frequency, the 2500 Series reference is variable. John Regazzi, the company’s CEO/president, commented, “The partitioning of the PLL frequency steps and reference tuning is carefully calculated. The object is to maintain a wide PLL bandwidth while providing sufficient suppression of spurious sidebands.”

A combination of techniques, including a fractional pre-scaler, keeps the division ratio N below a maximum of 512. Where division by N = 4 implies that the output frequency is 4x higher than the reference, a fractional division by N = 1/4 means the output is 4x lower than the reference. Figure 4 shows the phase noise performance of the 2500 Series Synthesizers.

Figure 4. Giga-tronics Model 2500 Phase Noise
Courtesy of Giga-tronics

Mr. Regazzi explained the curve’s shape. “Ideally, the phase-noise profile consists of two distinct sections. The first is a pedestal extending from the carrier to the loop bandwidth, followed by the noise of the free-running oscillator, which usually drops monotonically at 20 dB/octave to the noise floor. “

A PLL should reduce noise within its loop bandwidth, as Mr. Regazzi commented. On the other hand, the PLL output can only be as good as its reference input. So, it must be assumed that the -20-dB/octave slope for offsets less than about 1 kHz is a characteristic of the reference oscillator.

DDS-Based Synthesizers
Figure 5 shows a DDS-based generator. Key to its operation is the phase accumulator that provides an index into a look-up table. In operation, a programmable digital tuning word is repeatedly added to the accumulator total. Typically, the accumulator has several more bits than actually output to the look-up table. Having more bits supports very fine frequency resolution. If the sine value output from the look-up table and the following DAC have high resolution, the generated waveform will have good fidelity. Inevitably, truncation occurs because of limited look-up table size and DAC resolution, introducing distortion.

Figure 5. DDS Block Diagram
Courtesy of National Instruments

DDS techniques often are used in combination with PLL-based synthesizers. In contrast to the very high-frequency performance plotted in Figure 4, the phase noise performance of the Rohde & Schwarz Model SMB100A Signal Generator for outputs as low as 10 MHz is shown in Figure 6. The two figures give similar values for high-frequency operation at offsets between 10 kHz and 100 kHz. Either side of this band, the differences are interesting.

Figure 6. Rohde & Schwarz Model SMB100A Phase Noise
Courtesy of Rohde & Schwarz

For small offsets and low output frequencies, L_PM is as small as -80 dBc/Hz for the SMB100A. The datasheet states that a new DDS synthesizer was used in this generator for outputs from 9 kHz to 23.4375 MHz. So, Figure 6 is a good example of the very low L_PM achievable in a DDS-based design. Above 100-kHz offset, however, the Giga-tronics VCO would appear to have significantly lower noise, having rolled off to -140 dBc/Hz or less at a 1-MHz offset compared to -125 dBc/Hz for a 6-GHz output from the SMB100A.

The SMB100A also avoids the increase in phase noise at low frequencies that often accompanies synthesizer designs that use downconversion. When plotted as a function of output frequency, typical L_PM at 20 kHz offset remains below -145 dBc/Hz from 9 kHz to above 100 MHz.

Further Instrument Examples
In contrast to most synthesized generators, dBm’s Model SSG provides only a CW output from 10 MHz to 4 GHz. Nevertheless, the design emphasis has been on achieving good RF performance such as a -145-dBc noise floor and -107-dBc/Hz L_PM at 10-kHz offset from a 1-GHz carrier. The generator offers an optional 200-µs settling time, which is much faster than that of instruments using a slower YIG-based VCO.

The company’s Vice President of Sales Michael Cagney explained, “The Model SSG has been well-received because it is not a direct competitor to feature-laden, expensive analog and digital instruments that make up the bulk of the market. The SSG provides the capability to make basic measurements in a broad array of applications and is small, lightweight, and quick to set up.”

As Mr. Cagney noted, many more features usually are found in other manufacturers’ analog signal generators. The new Model MG37020A Microwave Signal Generator from Anritsu is no exception. However, it’s interesting to see which capabilities the various test and measurement suppliers emphasize to differentiate their products.

In the case of the MG37020A, 100-µs switching speed is a key headline. Leonard Dickstein, product marketing manager at the company, said, “The VCO-based architecture switches 10x faster than comparable YIG-based signal generators while maintaining low phase noise of typically -101 dBc/Hz at a 10- kHz offset from 10 GHz. The generator’s overall performance suits data-intensive applications such as antenna test, satellite payload test, and terrestrial microwave link testing.”

The MG37020A has a basic range of 2 GHz to 20 GHz. Several types of sweep modes are available, including linear, log, and list. In list mode, up to 10,001 nonsequential frequency-power sets can be stored and then addressed as a phase-locked step sweep. This means that every frequency step is phase locked, and the steps can be as small as the 0.001-Hz instrument resolution or as large as the full 20-GHz range.

Obviously, such a sweep could take a long time if each step required 1 or 2 milliseconds minimum. Nevertheless, some tests are not just about speed, so you also have the capability to dwell a variable amount of time from 50 µs to 30 s at each step.

For lower-frequency operation, a 10-MHz to 2.2-GHz option is available. Although this is a form of downconverter, it is a digital downconverter that uses successive divide-by-two circuitry. This approach claims to have reduced phase noise compared to heterodyne-based downconverters. The MG37020A is the latest in a range of synthesized analog generators from Anritsu.

Agilent Technologies also offers many product choices, each with different strengths. Although it’s not the newest generator, the Model E8663B is distinctive in its claim to have the lowest close-in phase noise of competing instruments. Coincident with this characteristic is a slow 7-ms nominal frequency switching speed that can be greater than 20 ms for steps to or from the instrument’s 3.2-GHz to 9-GHz upper frequency band.

If you need the absolute lowest close-in phase noise performance, perhaps for sensitive component testing, the E8663B provides a typical residual L_PM of -74 dBc/Hz at a 1-Hz offset for frequencies from 3.2 to 9 GHz. This value is about 35 dB lower than the comparable level of the Giga-tronics 2500, for example. At larger offsets, phase noise performance of the two instruments becomes comparable. If you were to compare curves of E8663B performance with those of Figure 4, for offsets less than 10 kHz, the E8663B curves would have about half the slope of the 2500, indicating a much lower phase noise reference.

Agilent’s John Hansen, senior product manager at the company’s Signal Sources division, discussed several aspects of synthesized generators. Both radar and communications testing involving high bits per hertz modulation such as 64 or 256 QAM are particularly sensitive to phase noise. These kinds of applications need low phase noise generators, but when that requirement is combined with high performance in other areas as well, higher cost inevitably results.

One of the more important generator capabilities is wideband modulation. Currently, Mr. Hansen said, this is attracting a great deal of attention. Spread-spectrum modulation expands the original signal bandwidth to occupy a much wider frequency range. CDMA originally required a 5-MHz bandwidth that was increased to 20 MHz by WCDMA and has exploded to greater than 500 MHz with the introduction of UWB.

A DDS approach operated at a high enough rate could be appropriate. One of the advantages of DDS-based synthesis is phase continuous frequency switching, an important modulation characteristic. And, DDS technology inherently can precisely change frequency as fast as a new tuning value can be programmed. These characteristics would seem to suit very wideband modulation requirements, provided the DDS circuitry can operate sufficiently fast.

A related direct digital approach is provided by high-speed arbitrary waveform generators (Arb). Arbs promise the ultimate flexibility in waveform time and amplitude manipulation, but their related RF performance is lacking.

A high-speed Arb provides a very convenient means of generating digitally modulated signals including pre/de-emphasis and even frequency hopping. The Tektronix Model AWG7102 with optional interleaved high bandwidth output has these and many other useful capabilities. This model features a 20-GHz clock rate and 5.8-GHz bandwidth.

However, in terms of phase noise, the AWG7102 specifies a maximum -85 dBc/Hz at 10-kHz offset from a 625-MHz carrier. This compares with -111 dBc/Hz for the Anritsu MG37020A with the low-frequency option, -130 dBc/Hz for the Agilent E8663B, -122 dBc/Hz for the Rohde & Schwarz SMB100A, approximately -120 dBc/Hz for the Giga-tronics 2500, and -107 dBc for the dBm Model SSG, all specified under similar conditions.

Summary
Today, you can’t have 100-dB dynamic range, 0.00001-Hz frequency resolution, very low phase noise, fast switching speed, and extremely wide modulation capabilities all from the same instrument. Nevertheless, for some applications, this is indeed the goal.

As the comparison of several analog generator models has shown, a wide range of architectures is available from which to choose. DDS and PLL designs have been discussed, but other approaches such as direct analog synthesis also are useful. This technique has extremely fast switching and low noise but is limited in the frequency step size it can provide at reasonable cost.

Selecting the best solution for your application is a matter of deciding which capabilities you must have and which are secondary. Certain combinations simply can’t be obtained. For example, several aspects of YIG-based designs are desirable, but YIG VCOs require about 1 ms to change frequencies. This is one reason that fast-switching generators use other types of VCOs.

Nevertheless, your range of choices continues to expand as new synthesized generators are introduced. Because the specifications are highly interrelated, don’t hesitate to ask for clarification if in doubt about claimed performance. And, evaluate short-listed models in your actual application before deciding which generator to buy. This always is the best proof that the proposed instrument really will perform as required.

References
1. Phase Noise Reference, Applied Radio Labs Design File, www.radiolab.com.au/designfile/pnref/pnref.htm.
2. Fox, A., “PLL Synthesizers,” Analog Dialogue, Analog Devices, 36-03, 2002.

November 2007