Recent advances in high-speed data converters now digitize or generate signals directly at radio frequencies up to several gigahertz, replacing traditional radio frequency (RF) components like mixers, local oscillators (LOs) and amplifiers. Thus arises the question of how to apply traditional RF device specifications like noise figure (NF), third-order intercept point (IP3) and phase noise. Although you can calculate these specifications, you must understand the limitations in applying the specifications to data converters, because they are not purely analog components.
Let’s start with IP3. IP3 assumes an input-to-output transfer function that follows a polynomial. See Equation 1:
The intermodulation product (IMD3) power increases in amplitude by the cube of the input power.
IP3 is the theoretical point at which the intermodulation product power equals the input power (Figure 1).
Now let’s look at the IMD3 behavior for the Texas Instruments ADC32RF45, a dual-channel 14-bit, 3GSPS analog-to-digital converter (ADC) capable of sampling frequencies above 3GHz. Figure 2 shows IMD3 vs. input amplitude at 1.78GHz and 2.949GSPS. Between the range of -8dBFS down to -14dBFS per tone, IMD3 follows a 3-to-1 slope with an input IP3 of 23dBFS. Given the full-scale input power of 6dBm, this translates to an input IP3 of 29dBm.
For an input level between -16 and -24dBFS, IMD3 flattens at ∼90dBFS before falling to -100dBFS below an input level of -24dBFS. The reason for this behavior is that the ADC has small mismatches that result in a nonlinearity that deviates from a simple polynomial around the amplitude of the pipeline stages.
IP3 also implies a gradual increase in intermodulation products with higher input power. For the ADC, however, once full scale is reached at -6dBFS per tone, the output will saturate, causing a rapid increase in IMD3 and higher-order intermodulation components.
ADC32RF45 IMD3 and IP3 model.
Consider the NF of an RF sampling ADC, which is the ratio of output noise to input noise. Although the input is analog and output digital, you can compare the noise when both are normalized to the ADC’s full scale. The ADC32RF45 lists the NF in the specifications table, but it is instructive to show how it is calculated from the ADC’s small-signal signal-to-noise ratio (SNR). In Table 1, we first calculate the output noise spectral density, then the input noise spectral density and then the noise figure.
It is important to use small-signal SNR rather than the SNR for a large signal, as the latter will include phase noise and energy from higher-order harmonics. The small-signal SNR is normalized to noise spectral density by dividing by the Nyquist bandwidth, resulting in a number with units (decibels to full scale/hertz). The input thermal noise is also normalized to full scale. The ratio (or subtraction in decibels) between the input and output noise spectral density provides the NF.
Calculation of NF from ADC SNR.
The last topic is phase noise. For an RF data converter, conversion clock phase noise has a similar effect as the local oscillator phase noise driving a mixer – the phase noise will modulate with large signals, potentially obscuring small signals nearby. The only difference for the clock phase noise is that the ratio of the input signal frequency to the clock frequency adjusts the sampled phase noise, as expressed by Equation 2:
Reference  offers a more detailed explanation and analysis.
Although RF sampling data converters are replacing traditional RF signal chains, you can still derive and use traditional RF device specifications as long as you recognize the limitations.
Stay tuned for the next Signal Chain Basics article, with advice on working with data converters, amplifiers, interface or other analog design challenges.