With the unprecedented proliferation of wireless communications, wireless engineering has become extremely complex in most applications, including cell phones, computer network components, satellite communications, and commercial broadcast transmitters, all of which require ultra-high stability and accuracy.

To ensure that communications products meet those requirements, multi-channel analog-to-digital converters (ADCs), also known as digitizers, are now widely used in R&D prototype development, manufacturing test, and field service testing of communication chips, sub-assemblies, and finished products.

PC-based digitizer cards have demonstrated significant advantages over oscilloscopes in terms of speed (sampling rate), resolution, and accuracy. But there’s often cause for confusion over manufacturers’ specifications, the numbers that describe the quality and performance range of each manufacturer’s digitizers.

While manufacturers publish critical performance metrics such as speed, memory, clock accuracy, and bandwidth, these specifications require further definition and should only be taken at face value.

That’s why it’s important to “look behind the numbers” to see what they may or may not mean and ensure that the digitizer you choose meets your particular requirements. Therefore, you may want to request additional information from the digitizer’s manufacturer to ensure that its model is right for your applications.

Flatness Of Frequency Response

The spectral “flatness” of the digitizer card’s frequency response is one of the important parameters that requires further clarification from the manufacturer. Flatness describes the frequency below which the frequency response curve remains within ±1 dB of 0-dB attenuation (Fig. 1). However, achieving flatness at high frequency (>100 MHz) with no distorting oscillations in the digitizer’s pass-band can be challenging.


1. Flatness, which is an important digitizer performance metric in communications, is defined as the frequency below which the frequency response remains within a ±1-dB band about 0 dB. The bandwidth is 350 MHz, but the curve is flat almost to 300 MHz.

Theoretically, you want the frequency response curve to be flat and sustained all the way to infinite frequency so the signal wouldn’t be attenuated at all. But in the real world, the acceptable attenuation is generally considered to be 3 dB.

The frequency below which the digitizer attenuates an input signal by less than 3 dB is called the cutoff frequency or bandwidth. The digitizer’s pass-band comprises frequencies that lie below this 3-dB frequency. To know the true pass-band of a digitizer, you can ask the manufacturer to provide the appropriate frequency response curve.

The curve in Figure 1 illustrates that while the signal has a bandwidth of 350 MHz, for example, the digitizer’s frequency response is flat only up to 300 MHz. In some cases, where the attenuation may be greater than 1 dB but less than 3 dB, the response may be oscillating rapidly at some frequencies.

This somewhat problematic situation would be indicated by the response curve quivering sporadically within the 0- to 300-MHz bandwidth. When attenuation exceeds 3 dB, flatness falls off significantly and noise occurs even more noticeably. In the absence of the full frequency response curve, the flatness provides much more information about the frequency response than just the bandwidth.

Since most applications use a combination of frequencies, a pass-band with insufficient flatness often will be detrimental to performance in wireless applications. So, it would be very helpful for users to request from their prospective suppliers some graphical or tabular references providing sample input signals (sine waves) and corresponding frequency response curves, rather than simply a bandwidth description. Don’t just ask about the bandwidth. Ask about the flatness. That’s the data you really need.

High Vertical Resolution

Digitizers should provide the best possible resolution, or signal fidelity. While almost all standalone digital oscilloscopes include embedded 8-bit digitizers, a 12-bit digitizer card enables much higher resolution: 4096 levels across the input range compared to only 256 levels attainable from an 8-bit oscilloscope. A more useful resolution is obtained with 16 bits. The 16X (65,536) levels allow smaller signal features to be detected in the time or frequency domain.

However, the nominal resolution stated by the digitizer’s manufacturer does not directly indicate its performance. The best single measure of performance is the effective number of bits (ENOB), which is the true resolution, allowing for the noise and distortion introduced by the digitizer instrument.

Generally, we can characterize the undesirable elements added to an input signal by a digitizer instrument as either noise or distortion. Of these two, distortion is almost always the dominant contributor to the degradation of signal fidelity at high signal frequency.

Once a signal has been distorted, there is no practical way to remove the distortion. Further, the signal-to-noise ratio (SNR) generally stays constant as a function of signal frequency, while the total harmonic distortion (THD) increases dramatically with signal frequency. Therefore, high vertical resolution and accompanying high ENOB and low THD near the intermediate frequency (IF) are important for good signal integrity.

The nominal resolution of a digitizer is theoretical, and the ENOB is what you actually get. For that reason, although the ENOB may not be stated in the product literature, users should look beyond the nominal resolution and get the ENOB from the digitizer manufacturer to determine if the vertical signal fidelity is sufficient for the application.

Precision Sampling

Compared to most digitizer applications, wireless communications applications have very strict requirements for the accuracy and stability of the sampling clock signal that determines the sampling rate of the digitizer’s ADC chips. Using a disciplined crystal oscillator, digitizers may achieve excellent sampling clock accuracy of about 1 part per million.

Nonetheless, communications applications routinely require even better clocking accuracy. Compared to standalone oscilloscopes, commercial digitizers provide much more ADC clocking flexibility. Digitizers often provide two methods of improving clock signal accuracy.

First, the user may provide an ultra-accurate external sampling clock signal, which is directly passed to the ADC chips. The external clocking input, which almost never is  available on standalone oscilloscopes, also allows the user to provide any sampling rate.

For example, a user designing communications receivers with built-in 887.3-Msample/s ADCs could simulate receiver operation with a commercial digitizer operating with an 887.3-MHz clocking signal being “fed” to the digitizer card from an external source.

The second method of improving clock accuracy requires a digitizer with a 10-MHz reference signal input. The signal is internally used to control a phase-locked loop (PLL) circuit that creates the digitizer’s sampling clock. For instance, the PLL would ensure that a 500-MHz sampling frequency is exactly 50 times the frequency of the input 10 MHz.

Not all digitizers can use either a selected external signal or a 10-MHz reference signal. But this ability is useful in many applications, allowing for the synchronization of multiple instruments, which can circumvent signal instability problems in many communications systems.

Deep Memory

In many wireless applications, communications engineers and manufacturers want to test signals or sine waves for durations ranging from several seconds up to many minutes or even hours. Typically, engineers want to acquire uninterrupted signals at high speeds like 500 Msamples/s for human time-scales of seconds or minutes.

In the past, this has required large amounts of dedicated on-board acquisition memory. Several Gsamples of on-board memory allow unbroken signal acquisition for several seconds at 500 Msamples/s.

Recently, data buses like PCI Express have emerged to allow continuous streaming of waveform data directly from the digitizer to fast hard-drive arrays (RAIDs) with terabytes of capacity. Digitizers on such platforms may stream data at 500 Msamples/s for a full hour or longer. Therefore, high digitizer memory or even streaming to high-capacity storage drives allow for the acquisition of communications signals for relatively long durations of up to hours.

Other aspects of digitizers also are important. But if you want to look at raw analog communications signals and characterize them accurately, you need to have a digitizer with appropriate flatness, vertical resolution, memory capacity, and sample rate. To make an informed selection of equipment using those criteria, users need to be sure that real-world information is available and request it if needed.

Figure 2 shows a typical PC digitizer. It plugs into a PCI Express or PCI bus slot on a computer to offer a wide range of specifications for selected applications. The front end has software control of input voltage ranges, coupling, and impedances. On-board sampling memory can be as large as 16 Gsamples/s in the PCI Express models.


2. The GaGe Razor CompuScope 16xx digitizer comes in four basic models with two or four simultaneous input channels. The sampling rate can be 100 Msamples/s or 200 Msamples/s. Input bandwidth can be 65 MHz or 125 MHz. Vertical resolution is 16 bits. Basic internal storage is 128 Msamples.