Function Generator

What is a Function Generator

A function generator is an electronic instrument that synthesizes repetitive or single-shot electrical waveforms with adjustable frequency, amplitude, offset, and shape. It belongs to the signal-source half of the test bench, the counterpart to the oscilloscope and the spectrum analyzer that observe a circuit. Where a DC power supply delivers steady energy, a function generator delivers a precisely controlled, time-varying voltage so the engineer can excite a filter, clock a logic input, drive a transducer, or stress a feedback loop and then measure how the device under test responds.

A complete instrument has four functional blocks: (1) a synthesis core that generates the digital waveform, today almost always a DDS phase accumulator with a lookup table or a point-by-point sample engine; (2) a digital-to-analog converter that turns the sample stream into an analog voltage; (3) a reconstruction filter and output amplifier that smooth and scale the signal; and (4) a 50 ohm output stage with programmable amplitude and DC offset. Around this core sit a reference clock for timing accuracy, a modulation engine for AM, FM, PM, FSK, and PWM, and a remote-control interface for automated test.

The vocabulary is worth pinning down because it appears inconsistently on purchase orders. A function generator classically produces a fixed menu of standard shapes. An arbitrary waveform generator (AWG) additionally replays any user-defined waveform stored point by point in sample memory. A signal generator usually refers to an RF source optimized for high-frequency carriers and modulation rather than baseband shapes. Because nearly every modern benchtop instrument does both function and arbitrary work, the practical distinctions are sample rate, arb memory depth, and vertical resolution, not the name printed on the bezel.

The history runs from analog to digital. Early function generators in the 1960s and 1970s built a triangle wave with an integrator driven by a switched current source, then shaped it into sine and square waves with a comparator and a diode shaper. Monolithic function generator chips such as the Intersil ICL8038 and the Exar XR2206 packaged this approach into a single device and defined the hobby and instrumentation bench for years. These analog designs were simple and cheap but suffered frequency stability around 0.1 percent per hour and limited amplitude accuracy.

Direct digital synthesis changed the field. By computing the waveform from a numeric phase rather than an analog ramp, DDS delivered frequency stability in the 500 ppm class and below, microhertz frequency setting resolution, and the ability to phase-lock channels and reference clocks. Refinements followed: Keysight Trueform blends DDS with a point-by-point architecture to suppress edge jitter, and Siglent TrueArb and EasyPulse target the same square and pulse fidelity problem. The frequency span of typical bench instruments runs from a few hertz to roughly 20 MHz for entry units and up to 120 MHz, 250 MHz at the high end, with dedicated RF generators handling higher carriers by other methods.

The application range explains why the category fragments into so many instruments. On the low-frequency side, a function generator drives the test of audio circuits, sensor signal conditioning, and mechanical and biomedical models that operate from fractions of a hertz upward, where the demand is amplitude accuracy and low distortion rather than speed. In the middle, the bulk of electronics development uses the 20 MHz to 120 MHz band to clock logic, exercise power-converter control loops, and sweep filters. At the top, 250 MHz arbitrary work feeds serial-data, communications, and radar baseband emulation. Because no single instrument optimizes all three, procurement engineers match a specific instrument class to a defined band, a defined waveform, and a defined load rather than buying a notional universal source. On the bench it rarely works alone: the response it stimulates is read back with a digital multimeter for static values, and a programmable electronic load often completes a source-and-sink test loop for power and converter work.

Waveform Types and Output Modes

Selection often starts with the waveform you need to produce. Standard functions cover the majority of bench work, but the output mode (continuous, swept, gated, burst, or modulated) frequently decides which instrument is adequate. The table below summarizes the standard waveform set, the parameter that usually limits it, and where it is used.

Sine is the reference waveform for frequency-domain testing, and its quality is judged by harmonic distortion and amplitude flatness across the band rather than by shape alone. Square waves drive logic and switching circuits; their value lies in adjustable duty cycle and clean, fast edges, which is exactly where pure DDS struggles. Triangle and ramp share a linear slope and are distinguished by symmetry: a symmetric triangle rises and falls equally, while a ramp is asymmetric, approaching a sawtooth as symmetry tends toward 0 or 100 percent.

Pulse output is a discipline of its own: independent control of period, width, and edge rate lets the instrument emulate clocks, strobes, and timing signals. Noise provides broadband stochastic stimulus for signal-to-noise and dither testing, and DC output turns the same channel into a programmable bias source. Arbitrary playback is the open-ended mode: load a digitized waveform, a captured glitch, or a synthesized communications burst, and the instrument replays it sample by sample at the chosen sample rate.

Beyond the shape, four output modes recur on every spec sheet. Continuous runs the waveform indefinitely. Sweep walks the frequency between two limits, either linearly or logarithmically, to trace a frequency response in one pass. Burst emits a defined number of cycles on a trigger, useful for gated stimulus and settling tests. Modulation impresses one signal onto another. The common modulation types are AM, FM, PM, FSK, and PWM, which let a single channel emulate communications envelopes, frequency-shift keying, and pulse-width-modulated control signals for system-level test.

Channel count interacts with these modes. A one-channel instrument suffices for a single stimulus, but two-channel models can produce independent or coupled signals, hold a fixed phase relationship between channels, and use one channel to modulate or trigger the other. Tracking and coupling features, where the second channel mirrors or phase-locks to the first, are common selection criteria for I/Q, differential, and clock-plus-data test setups.

Synthesis Architectures Compared

Two facts drive almost every architecture trade-off: how the instrument computes each sample, and how cleanly it turns samples into a continuous voltage. Three families dominate the benchtop market today, plus a legacy analog approach. The table compares their engineering character.

Direct digital synthesis is the baseline. A fixed reference clock drives a phase accumulator: each clock cycle adds a tuning word M to a running phase register of width N (commonly 24 to 48 bits). The high bits of the phase address a waveform lookup table, and the table value feeds the DAC. The output frequency follows the relation fout = M times fclk divided by 2 to the power N. Because N is large, the smallest frequency step is fclk divided by 2 to the power N, which lands in the microhertz region. This is why DDS instruments set frequency with extraordinary precision and lock cleanly to an external reference.

DDS has one well-known weakness. To raise the output frequency the accumulator skips table entries, and the phase value rarely lands exactly on a sample boundary. The address is truncated, so the moment an edge actually appears is quantized to the clock period. On a sine wave this is invisible, but on a square or pulse wave it shows up as edge jitter of a few hundred picoseconds, which matters for timing and clock-recovery testing. Reconstruction filtering also imposes a Nyquist limit: usable signal bandwidth stays well below half the sample rate, and engineers keep the sample rate roughly 5 to 10 times the highest signal component.

Point-by-point and blended architectures address the jitter problem directly. Keysight Trueform technology blends DDS frequency agility with a point-by-point playback path so that edges are placed without truncation jitter, reaching the sub-1-picosecond and 0.03 percent total harmonic distortion class on the 33500B and 33600A series. Siglent answers the same need with TrueArb, a point-by-point arb engine, and EasyPulse, a dedicated pulse path that produces low-jitter square and pulse signals (the SDG1000X reaches roughly 17 ns rise and fall times with pulse frequency to 12.5 MHz). Rigol SiFi targets fidelity at low frequencies on the DG1000Z. The lesson is that the synthesis label predicts pulse quality more reliably than headline bandwidth does.

Two memory-related specs complete the architecture picture. Sample rate sets how finely arbitrary waveforms are reconstructed in time, ranging from 150 MSa/s and 200 MSa/s on mid instruments to 1.2 GSa/s, 2 GSa/s at the high end. Arb memory depth sets how long a unique waveform can be before it must repeat, from a few megasamples on bench units to 16 Msample and 64 Msample on high-end models. A long, complex waveform such as a modulated communications frame needs both high sample rate and deep memory; a simple repetitive arb needs neither.

Output Stage and Interface Standards

Two practical issues sink more bench measurements than synthesis quality ever does: output termination and remote-control compatibility. Both are governed by conventions every engineer should internalize before reading amplitude numbers off a spec sheet.

The output stage of a function generator presents a 50 ohm source impedance, chosen to match the 50 ohm coaxial cable that interconnects bench instruments. The amplitude the instrument displays assumes the output is terminated into a matching 50 ohm load, which forms a 2-to-1 voltage divider with the internal resistance. Connect a high impedance load instead, such as a 1 megaohm oscilloscope input, and no division occurs: the measured voltage is twice the programmed value. This is the single most common bench surprise. The cure is to set the output termination correctly, High-Z for a scope or 50 ohm for a matched line, so the instrument rescales internal amplitude and the displayed value matches the measurement.

Termination also governs signal integrity. Fast edges reflect off any impedance discontinuity, so a square or pulse line should be terminated in 50 ohm at the receiving end and kept short. A pulse with sub-nanosecond edges into an unmatched load will ring and overshoot regardless of how clean the generator output is. For high-amplitude or high-current drive the 50 ohm output stage is current limited; the correct path is an external wideband amplifier, not pushing the generator beyond its rated output window.

On the control side, automation depends on a stack of standards layered over decades. The table summarizes the remote-control conventions that appear in instrument specifications.

SCPI (Standard Commands for Programmable Instruments), first released in 1990, defines a common text command syntax so that, for example, an amplitude command reads the same across compliant instruments. SCPI sits on top of IEEE-488.2, which defines protocol and data format, which in turn rode the IEEE-488 (GPIB) hardware bus that began life as Hewlett-Packard's HP-IB in 1965. Modern instruments carry the same SCPI commands over newer transports: USBTMC makes a USB device behave like a GPIB instrument, and LXI carries control over standard Ethernet, the common choice for distributed test racks.

For mixed-vendor automation, IVI (Interchangeable Virtual Instruments) drivers from the IVI Foundation let instruments of the same class share test code, so a function generator can be swapped for a different brand with minimal software change. When specifying for an automated test system, confirm SCPI compliance, the set of available transports (GPIB, USB, LAN), a published VISA address scheme, and whether an IVI driver exists. These determine integration effort far more than front-panel ergonomics.

One further interface concerns timing rather than commands. The reference clock input and output, usually a 10 MHz BNC connector, let several instruments share a common timebase so their frequencies and phases stay coherent. Lock a function generator, an oscilloscope, and a frequency counter to one reference and their measurements no longer drift relative to each other, which is essential for phase, jitter, and beat-frequency work. Two-channel instruments extend the same idea internally with phase coupling between channels, and many add a trigger input and a sync or marker output so the generator can start a burst on an external event or flag a point in its own waveform to the rest of the bench. When timing relationships matter, check for the 10 MHz reference, the trigger and sync lines, and the channel phase-coupling specification, not just the headline output band.

Key Specification Parameters

A function generator datasheet can list dozens of lines, but a handful of parameters drive nearly every selection. The first comparison table puts the headline specs of representative instruments side by side; the prose then decodes what each parameter actually means in use.

Maximum frequency is the headline bandwidth, but it applies to the sine output; square, pulse, and arbitrary signals are usable to a lower frequency because they need more harmonic content and reconstruction headroom. Read the spec per waveform, not as a single number. Frequency resolution on DDS instruments reaches the microhertz region (the smallest step is the reference clock divided by 2 to the accumulator width), which matters for beat-frequency, Doppler, and precise sweep work.

Sample rate and vertical resolution are the two arbitrary-mode specs. Sample rate, from 150 MSa/s to 2 GSa/s across the market, sets time-domain granularity; keep it 5 to 10 times the highest signal component for clean reconstruction. Vertical resolution sets amplitude granularity: 14 bits give 16,384 steps and suit most testing, while 16 bits give 65,536 steps and lower quantization noise for low-distortion and sensor-emulation work. Arb memory depth caps how long a unique waveform can run before repeating, from a few Msample to 16 Msample and 64 Msample on high-end units.

Output amplitude typically spans about 1 mVpp to 10 Vpp into 50 ohm, doubling to roughly 20 Vpp into a high impedance load. The low end near 1 mVpp matters for low-power testing, where externally attenuating a large signal would add noise. DC offset is adjustable across a similar window, constrained so that signal peak plus offset stays inside the total output range. Signal purity appears as total harmonic distortion (down to roughly 0.03 percent on the best benchtop units), spurious-free dynamic range, and sine flatness across the band.

Edge jitter and rise / fall time describe pulse fidelity. High-end point-by-point instruments quote intrinsic jitter under 1 ps, while DDS-based units carry a few hundred picoseconds; mid units reach roughly 17 ns rise and fall times. Channel count (one or two) and inter-channel phase coupling decide whether the instrument can produce differential, I/Q, or clock-plus-data pairs. Finally, modulation depth and the supported types (AM, FM, PM, FSK, PWM), plus sweep and burst modes, complete the functional spec. The table below maps common test tasks to the parameter that should drive the choice.

Selection Decision Factors

To turn the preceding chapters into a specific model, work through the decision sequence below in order. Most selection mistakes come not from a single wrong spec but from deciding bandwidth before defining the waveform and the load, then discovering the pulse fidelity or interface is wrong.

1. Define the waveform and output mode first: Standard functions, arbitrary playback, or pulse-critical signals. If you need clean square and pulse edges, prioritize a point-by-point or dedicated pulse engine (Trueform, TrueArb and EasyPulse) over raw DDS bandwidth.
2. Set the bandwidth from the signal, not the headline: Read maximum frequency per waveform. A 120 MHz sine rating does not mean a 120 MHz clean square wave. Allow harmonic and reconstruction headroom of several times the highest component.
3. Choose sample rate and vertical resolution for arbitrary work: Keep sample rate 5 to 10 times the fastest signal component, and pick 16-bit resolution for low-distortion or sensor-emulation tasks, 14-bit for general stimulus.
4. Size the arb memory depth: Long, non-repeating waveforms (communications frames, captured transients) need megasamples to tens of megasamples; simple repetitive arbs do not.
5. Fix the output requirements: Confirm amplitude range into your actual load (50 ohm vs High-Z), minimum amplitude for low-power work, DC offset window, and whether an external amplifier is needed for drive.
6. Decide channel count and coupling: One channel for single stimulus; two channels with phase coupling for differential, I/Q, or clock-plus-data pairs.
7. Specify modulation and sweep needs: Confirm the required modulation types (AM, FM, PM, FSK, PWM), sweep linearity (linear or logarithmic), and burst control if your test plan uses them.
8. Lock the interface and automation path: For bench-only use the front panel suffices; for an automated rack confirm SCPI compliance, transports (GPIB, USB / USBTMC, LAN / LXI), a VISA address, and an IVI driver for brand interchangeability.

One last dimension is commonly overlooked: serviceability and calibration support. A function generator is a metrology source, so its amplitude and frequency accuracy must remain traceable over a multi-year service life, verified periodically against a reference instrument or a signal calibrator. Confirm the manufacturer offers calibration service and a documented calibration interval, that firmware is updatable, and that local support and spare parts exist. Keysight, Tektronix, Siglent, and Rigol all maintain regional service in major markets, which matters when an instrument anchors a production test line for 5 to 10 years. A budget unit that cannot be recalibrated locally becomes a hidden cost the first time it drifts out of tolerance.