Direct Digital Synthesis

Direct digital synthesis — DDS — is one of the most elegant ideas in modern radio electronics. Instead of building an analog oscillator and fighting its tendency to drift, or building a PLL and wrestling with loop dynamics, DDS generates a waveform entirely in the digital domain and then converts it to an analog signal. The result is a signal source with sub-hertz frequency resolution, instantaneous frequency switching, programmable phase, excellent stability derived from a crystal clock, and remarkable flexibility — all from a single compact IC.

DDS appears in equipment ranging from hobby modules costing a few dollars to military frequency-hopping radios. Every modern function and arbitrary waveform generator is built around it, and chips such as the AD9833, AD9850 and AD9951 from Analog Devices have powered countless signal generators, sensor exciters and radio projects. Understanding DDS at the circuit and mathematical level will let you design, troubleshoot and evaluate any equipment that uses it.

The Core Concept: Counting Through a Circle

Imagine a circle representing one complete cycle of a sine wave — 360 degrees of phase, from 0° to 360°. A pointer rotates around this circle at a constant angular velocity. Every time the pointer completes a full revolution, one cycle of the sine wave is generated. The sine wave amplitude at any moment is the sine of the current pointer angle. If the pointer spins once per second, the output frequency is 1 Hz. If it spins 14 million times per second, the output is 14 MHz.

Now replace the continuously rotating pointer with a digital counter. At every clock tick, the counter advances by a fixed step size called the phase increment (also called the frequency tuning word). When the counter overflows — counting past its maximum value and wrapping back to zero — that represents one complete revolution of the circle, one complete cycle of the output sine wave.

The output frequency depends on two things: how fast the counter ticks (the clock frequency) and how far it advances with each tick (the phase increment). A large phase increment means the counter wraps around quickly — high output frequency. A small phase increment means it wraps slowly — low output frequency. This is the entire conceptual basis of DDS.

DDS Architecture: Three Essential Blocks

DDS architecture: the frequency tuning word programs the phase increment added to the phase accumulator on each clock cycle. The accumulator output addresses a sine lookup table (ROM), whose output drives a DAC. The analog DAC output passes through a low-pass reconstruction filter to remove clock artifacts and produce a clean sine wave.

View Larger

Every DDS system has three essential blocks between the crystal clock input and the analog output.

1. Phase Accumulator

The phase accumulator is an N-bit binary counter. At every rising edge of the clock, it adds the frequency tuning word (FTW) to its current value. When the accumulated value exceeds 2^N, the counter overflows and wraps back to zero. This overflow represents one complete cycle of the output waveform.

The most significant bits of the accumulator output are used to address the lookup table. If the accumulator is 32 bits wide but the lookup table has only 12-bit addressing, the top 12 bits of the accumulator select the table entry. The bottom 20 bits are simply discarded — they provide sub-cycle frequency resolution below the lookup table granularity.

2. Sine Lookup Table (Phase-to-Amplitude Converter)

The lookup table (implemented as ROM inside the DDS IC) maps each phase value to the corresponding sine wave amplitude. For a 12-bit addressed table with 4,096 entries, each entry stores the 14-bit or 16-bit digital value representing sin(2π × address / 4096). When the accumulator outputs address 0, the table outputs 0 (sin 0° = 0). At address 1024, the table outputs its maximum positive value (sin 90° = 1). At 2048, it outputs 0 again (sin 180°). At 3072, it outputs its maximum negative value (sin 270° = −1).

The table is typically stored in ROM and is implemented very efficiently in silicon. It only needs to store one quadrant (0° to 90°) using symmetry — the other three quadrants are generated by sign inversions.

3. Digital-to-Analog Converter (DAC)

The DAC converts the digital amplitude values from the lookup table to an analog voltage or current. Its output is a staircase waveform — the sine wave sampled at the clock rate, with sharp transitions between each sample. After the DAC comes the reconstruction filter (a low-pass filter) that smooths these staircase steps into a continuous sine wave.

The Phase Accumulator in Detail

Consider a 32-bit phase accumulator with clock frequency F_clk. The accumulator counts from 0 to 2³² − 1 = 4,294,967,295. If the frequency tuning word (FTW) is 1, the accumulator advances by 1 per clock cycle. It takes 4,294,967,296 clock cycles to overflow and complete one output cycle. With a 1 GHz clock, the output frequency would be 1,000,000,000 / 4,294,967,296 ≈ 0.233 Hz. That is the minimum frequency step (frequency resolution) for this system.

If the FTW is set to 429,497 (approximately 2³² / 10,000), the accumulator overflows approximately 10,000 times per second for every gigahertz of clock frequency — actually, it overflows at exactly 429,497 / 4,294,967,296 × F_clk Hz, which is about 0.0001 × F_clk.

The general formula for the output frequency is derived directly from this counting process:

The DDS Frequency Formula

Let us examine this formula carefully because it reveals everything important about DDS behavior.

F_out ∝ FTW: Output frequency is directly proportional to the tuning word. Double the FTW, double the frequency. This makes DDS extremely easy to program — no loop dynamics, no lock time, no analog tuning voltage. Just write a number and the frequency changes instantly.

F_out ∝ F_clk: If the clock changes, all output frequencies change by the same ratio. If the clock drifts by 1 ppm, the output drifts by 1 ppm at all frequencies. The DDS output is exactly as stable as its clock — no more, no less.

F_out ∝ 1/2^N: A wider accumulator gives finer frequency steps. Doubling N (adding one bit) halves the minimum frequency step without changing the clock or tuning word.

Maximum output frequency: Limited by the Nyquist theorem to F_clk/2. In practice, the reconstruction filter rolls off before Nyquist, so the usable output range is typically below 40% of F_clk to maintain good spurious-free performance.

DDS Output Frequency Calculator

Worked Examples

Frequency Resolution

The frequency resolution of a DDS system — the smallest frequency step it can make — is determined entirely by the accumulator bit width and the clock frequency:

This extraordinary frequency resolution is DDS's greatest advantage over PLL synthesizers. An integer-N PLL with 10 kHz comparison frequency can only tune in 10 kHz steps. A DDS system with a 32-bit accumulator can tune in steps of 0.03 Hz — 300,000 times finer — while maintaining the same clock stability. This is why DDS is used for fine tuning in many signal generators and radios, providing continuous smooth tuning while a PLL handles coarse frequency selection.

Adding more bits to the accumulator increases resolution at no cost to the maximum output frequency or the noise floor. This is why modern DDS ICs use 24-, 32-, 40-, or even 48-bit accumulators. The silicon cost of extra flip-flops is minimal.

The Nyquist Limit and Why Output Must Be Below Half Clock

The Nyquist-Shannon sampling theorem states that a sampled system cannot accurately represent frequencies above half the sampling rate. In DDS, the clock is the sampling rate and the output is the sampled signal. If you try to set the output frequency above F_clk/2 (the Nyquist frequency), the output aliases — it folds back to a different, incorrect frequency.

Aliasing example: with a 100 MHz clock and output set to 60 MHz, the DDS actually produces a signal at 100 − 60 = 40 MHz (its alias). The Nyquist limit is absolute: there is no way to produce a clean output above F_clk/2 with a standard DDS architecture. This is why DDS ICs always specify the clock frequency separately from the maximum output frequency — the maximum output is always less than F_clk/2.

In practice, the reconstruction filter limits the usable output to about 40% of F_clk because the filter cannot achieve a perfectly sharp cutoff right at Nyquist. A 100 MHz clock gives reliable, clean outputs up to about 40 MHz. For higher frequencies, you either use a faster clock or follow the DDS with a multiplier stage.

Spurious Outputs and SFDR

DDS output spectrum showing the desired output carrier and the spurious signals that determine SFDR. Phase truncation spurs appear at predictable offsets from the carrier. Harmonic spurs appear at multiples of the output frequency. The reconstruction filter suppresses clock-rate images but not close-in phase truncation spurs.

View Larger

Unlike a sine wave generator, a DDS does not produce a perfectly pure output. Several mechanisms create unwanted spurious signals:

Phase Truncation Spurs

The phase accumulator may be 32 bits wide, but the sine lookup table may only use the top 12 bits for addressing. The lower 20 bits are discarded — this is called phase truncation. The phase error introduced by truncation is periodic and creates spurs at predictable frequencies relative to the output carrier. Phase truncation is the dominant source of close-in spurious signals in DDS.

The worst-case spur from phase truncation is approximately −6 dBc × (N − B), where N is the accumulator width and B is the address bits used for the lookup table. For N = 32 and B = 12, worst-case spur ≈ −6 × (32−12) = −120 dBc — excellent. In practice, the actual worst-case SFDR (spurious-free dynamic range) is around −80 to −100 dBc due to DAC and circuit non-idealities.

DAC Non-Linearity Spurs

A real DAC is not perfectly linear. Minor errors in each bit weight create harmonic distortion that appears as spurious signals at multiples of the output frequency. These decrease in level with more bits and better DAC architecture. The AD9850's 10-bit DAC gives SFDR of about −50 dBc under typical conditions. High-end DDS ICs with 14- or 16-bit DACs achieve SFDR of −80 to −100 dBc.

Clock Feedthrough

Some of the clock signal couples to the DDS output through capacitive and substrate paths inside the IC. This appears as a spur at exactly the clock frequency and its harmonics. Good PCB shielding and decoupling reduce but cannot eliminate this. The reconstruction filter attenuates clock-rate spurs significantly since they are near or above F_clk.

Spurious-Free Dynamic Range (SFDR)

SFDR is the ratio of the output carrier level to the strongest spurious signal, expressed in dBc (decibels below carrier). It is the key specification for DDS cleanliness. A DDS with SFDR of −70 dBc has its worst spur 70 dB below the carrier. That is ample for most test signals, and for anything that transmits it comfortably meets typical regulatory limits on spurious emissions (40–60 dB below the carrier), especially after the output filter provides additional suppression.

The DAC and Reconstruction Filter

The DAC output is a staircase waveform — the sine values sampled at F_clk. This staircase contains the desired output frequency plus images of it centered around the clock frequency (at F_clk ± F_out) and at all higher multiples. A reconstruction low-pass filter removes these images and the clock frequency, leaving only the desired sine wave.

The filter cutoff must be placed above the maximum desired output frequency (to pass the signal) but well below the clock frequency (to reject the first image at F_clk − F_out). For a 125 MHz clock with maximum output of 50 MHz, a cutoff around 60–70 MHz works well. Fifth- or seventh-order elliptic or Chebyshev filters are commonly used because they provide steep attenuation above the cutoff.

In a fixed-frequency application (such as a 40 kHz ultrasonic transducer driver, or a single-frequency test source), the reconstruction filter can be a narrow band-pass filter centered on the output frequency, providing excellent harmonic suppression. In a wide-tuning signal generator, an elliptic low-pass filter just below half the clock frequency removes the images while passing the whole output range.

Phase and Frequency Modulation with DDS

DDS has a unique capability that analog oscillators and PLLs lack: the ability to change not just frequency but also the phase of the output signal, instantly and by a precisely controlled amount. This is because the phase accumulator value at any moment completely defines the output phase. Changing the accumulator by a fixed offset immediately shifts the output phase by that fraction of a full cycle.

This makes DDS ideal for phase-shift keying (PSK), used in data links from RFID tags to satellite modems. A microcontroller can command the DDS phase register directly, producing instantaneous, clean phase transitions without the amplitude glitches that can occur with analog switching methods.

For continuous-phase frequency shift keying (CPFSK) — used in many data links, modems and remote-control systems — the DDS simply changes the FTW from one value to another. Because the accumulator state is preserved through the FTW change, the output phase is continuous through the frequency transition. No phase discontinuities, no transients, no switching glitches. This is important for minimizing the bandwidth occupied by a data signal.

DDS Applications

Tunable Signal Sources for Radios

DDS is nearly ideal as a tunable signal source for receivers and transmitters. A 32-bit DDS with a 125 MHz clock provides 0.029 Hz resolution across the whole range from audio to about 50 MHz with the stability of the clock oscillator. There is no warm-up drift, no microphonics, no LC tank to adjust. The microcontroller simply writes a new FTW to the DDS IC over SPI, and the frequency changes in nanoseconds. Many modern radios use DDS or DDS-like fractional-N clock chips (such as the Si5351) for exactly these reasons.

Signal Generators and Frequency Standards

Laboratory signal generators from Rigol, Siglent, Keysight and BK Precision use DDS as their core frequency generation engine. A 200 MHz clock with a 32-bit or 48-bit accumulator provides coverage from DC to 80+ MHz with millihertz resolution and excellent frequency accuracy — directly tied to the internal 10 MHz TCXO or an external reference. Arbitrary waveform generators extend the idea by replacing the sine lookup table with any waveform stored in memory. For hobby use, low-cost DDS modules make inexpensive signal generators for testing and alignment.

Sensor Excitation and Frequency Sweeps

Many sensors are measured by exciting them with a precise AC signal and measuring the response. Impedance analyzer chips such as the AD5933 contain a DDS that sweeps the excitation frequency over a range while an ADC measures the response — used for corrosion monitoring, biological impedance measurement and material testing. Ultrasonic flow meters and proximity sensors, eddy-current sensors and LVDT position sensors use DDS to generate their excitation signals, where the precise frequency and phase control lets the system detect tiny changes in the response.

Digital Receivers and Local Oscillators

Digital receivers and software-defined instruments use DDS techniques internally for quadrature mixing and frequency generation. The "digital down-conversion" inside a digital receiver involves multiplying the incoming samples by a numerically controlled oscillator (NCO) — which is essentially the accumulator and sine table part of DDS without the DAC step. This allows the receiver to tune to any frequency within its bandwidth with sub-hertz precision by simply changing the NCO FTW.

DDS vs PLL Synthesizer: When to Use Which

Many modern radios and instruments combine both: a DDS where its sub-hertz resolution and fast switching shine, and a PLL for very high frequency local oscillators where its superior phase noise and high-frequency capability matter. The combination exploits the strengths of each technology while avoiding their weaknesses.