Cellular networks today provide significant data and video in addition to traditional voice at faster rates than ever before. This has led to new modulation techniques and new air interface standards that rely on complex digital techniques.
While digital techniques enable systems to run faster and consume less power using smaller package sizes and at higher reliability with each generation, these systems place new demands on the RF and analog signal acquisition portion of the system. The complex modulation and wide bandwidth of base station transmitters result in higher crest factors for the power amplifier (PA).
To meet the more stringent requirements in the presence of higher crest factors, the PA is typically oversized to enable operation in the linear region. Without digital correction, the PA efficiency can be on the order of 10 percent – meaning 200W is required to run a 20W PA.
The PA is the largest consumer of electrical power in a base station and therefore a significant factor in the operating expense for the cellular service provider. To improve PA efficiency, digital techniques are used for crest factor reduction (CFR) and digital predistortion (DPD).
While an amplifier is most efficient when driven into saturation, it is also highly nonlinear in saturation. Complex digital modulation requires extremely high linearity from the PA, which consequently means that it must be driven well below saturation.
Operating the PA just below saturation offers good efficiency if there is a way to compensate for the inherent nonlinearity of the amplifier. Digital predistortion has emerged as the preferred method of PA linearization.
DPD is a feedback technique where the output of the PA is sampled and converted to digital data. A distortion-free transmit signal stored in a FIFO is compared to the feedback signal, creating an inverse of the transfer function.
This is summed with the transmit data after CFR to reduce the nonlinearity of the PA output. An adaptive algorithm or a lookup table may be used to produce the compensating digital signal or the two methods can be combined, but this is beyond the scope of this article. Here, we will focus on the analog requirements for the receiver that samples the PA output.
The receiver is the signal chain that converts from RF to digital (Figure 1 below ). Key design requirements are the input frequency range and power level, the intermediate frequency, and the bandwidth to be digitized. Some of these are derived directly from the PA speci- fications and some are optimized at design time.
|Figure 1: The receiver is the signal chain that converts from RF to digital|
The baseband transmit signal is upconverted to the carrier frequency and is defined in frequency by the air interface standard: W-CDMA, TD-SCDMA, CDMA2000, LTE etc. Therefore, the output spectrum to be sampled exists in a well-defined range of frequencies, the desired channel.
Since the purpose of the DPD loop is to measure the PA transfer function, it is not necessary to separate the carriers in multicarrier systems or demodulate the digital data. It is only necessary to capture information regarding the entire desired channel.
PA nonlinearity produces odd order intermodulation products, which constitute spectral regrowth in the adjacent and alternate channels. By definition, third-order products appear at 2fa + fb, 2fb + fa, 2fa “fb and 2fb “fa, where fa and fb are the frequencies of two signals within the desired channel that cause intermodulation distortion outside the channel.
For a modulated channel, third-order products appear within a range of three times the bandwidth of the desired channel (Figure 2 below ).
|Figure 2: For a modulated channel, third-order products appear within a range of three times the bandwidth of the desired channel.|
Likewise, fifth-order products appear within a range of five times the bandwidth and seventh-order products within seven times the bandwidth. Therefore, the DPD receiver must acquire a multiple of the transmit bandwidth equivalent to the order of the intermodulation products being linearized.
The trend in current development is to mix the desired channel to an intermediate frequency (IF) and capture the full bandwidth of all the intermodulation products. The exact IF is chosen to ease filtering and avoid other frequencies that are already fixed based on specification requirements.
The sample rate is similarly chosen as a multiple of the digital modulation chip rate, for example, 3.84MHz in W-CDMA. Finally, the Nyquist theorem dictates that the sample rate must be at least twice the sampled bandwidth.
Although many configurations are acceptable, one that meets these constraints is an IF of 184.32MHz, an ADC sample rate of 245.76MHz and a bandwidth of 122.88MHz. In the case of a 20W PA, the average output power is 43dBm.
The peak to average ratio (PAR) is about 15dBm. To set the average input power to the mixer of the receive chain at -15dBm, the combination of the coupler and attenuator insertion loss needs to be 58dB (See Figure 1, earlier ). The in-band noise of the PA is specified by the WCDMA standard at a maximum of -13dBm/MHz (-73dBm/Hz).
Therefore, the combination of the coupler and attenuation (-58dB) and the PA noise limit (-13dBm/MHz) yields a receiver sensitivity level that must be below -71dBm/MHz (-131dBm/Hz). For sufficient margin, a number at least 6dB to 10dB better than this is desirable. This sets the frequency plan, power level and sensitivity requirements for the DPD receiver.
Integrated DPD receiver
Once the system requirements are defined, the task turns to the circuit implementation using a mixer, IF amplifier, ADC, passive filtering, matching networks and supply bypassing.
While calculations and simulations are helpful, there is no substitute for evaluation of real hardware, which generally leads to multiple PCB iterations. However, a new class of integrated receivers based on system in package (SiP) technology simplifies this task.
The LTM9003 digital predistortion Module receiver is a fully integrated DPD receiver, essentially RF-to-bits. The microModule technology uses a thin, multilayer laminate substrate made of bismaleimide triazine (BT) material.
The multilayer substrate allows the design of complex circuits using RF components, standard wire-bonded IC die and traditional passive components. The circuit is encapsulated with standard IC package molding compound and the land grid array (LGA) pad arrangement is consistent with current surface mount assembly methods.
The result is a subsystem that looks and feels like a traditional IC, is thoroughly tested, ensuring the high reliability of an IC, yet integrates components from different semiconductor process technologies along with passives in a compact footprint smaller than traditional implementations.
The LTM9003 consists of a high linearity active mixer, an IF amplifier, an L-C bandpass filter and a high-speed ADC (Figure 3 below ).
|Figure 3: The LTM9003 consists of a high linearity active mixer, an IF amplifier, an L-C bandpass filter and a high-speed ADC.|
The wire-bonded bare die assembly ensures that the overall form factor is highly compact, but also allows the reference and supply bypass capacitors to be placed closer to the die than possible with traditional packaging.
This reduces the potential for noise to degrade the fidelity of the ADC. This idea extends to the high frequency layout techniques used throughout the receiver chain of the LTM9003.
The integration eliminates many challenges of driving highspeed ADCs. Linear circuit analysis cannot account for the current pulses resulting from the sample-and-hold switching action of the ADC.
Traditional circuit layout requires multiple iterations to define an input network that absorbs these pulses, is absorptive out of band, yet works seamlessly with the preceding amplifier. The IF amplifier must also be capable of driving this network without adding distortion. Solving these challenges may be the greatest hidden attribute of the LTM9003 microModule receiver.
The passive bandpass filter is a third order filter with an extremely flat passband. Within the center 25MHz of the band it exhibits less than 0.1dB ripple and over the entire 122MHz the passband ripple is only 0.5dB. The third order configuration ensures that the shoulders of the frequency response are monotonic, which is important for many DPD algorithms.
The overall performance of the LTM9003 greatly exceeds the system requirements described above. With a single tone at – 2.5dBm, which is equivalent to -1dBFS at the ADC, the SNR is typically -145dBm/Hz.
This figure is well below the target value of -131dBm/Hz defined by the W-CDMA standard. The worst case harmonics are 60dBc. The IIP3 figure of 25.7dBm means that the LTM9003 could support an ACPR of 87dBc if the PA were linear enough. Relative to the system requirements and the capability of the best PAs available, the LTM9003 greatly exceeds the requirements. The entire chain consumes about 1.5W from a 3.3V and a 2.5V supply, yet requires a circuit board area of only 11.25mm x 15mm.
Traditional high integrated ICs may offer flexibility in terms of programmable modes or selectable features, but this adds complexity and often some loss of performance. By changing the values of the passive components or substituting ICs that are optimized as a group, the LTM9003 is available in specific versions, with no loss of performance or increased complexity.
At a high level, DPD allows you to run the PA with less backoff. The result is that the PA is more efficient and therefore consumes less power for the same output power level. As discussed earlier, the PA is the most significant factor in electrical power consumption for the base station.
If your company has a “green initiative,” DPD is your contribution. Regardless, using less electricity reduces operating expenses for the service provider, thus making your product more competitive.
At the board level, the Module packaging integrates all of the key components into a small area including the passives for filtering and decoupling. This saves board space, simplifies layout and makes room for other features that further increase the value of the product.
At the engineering level, the LTM9003 saves time. Filter design and component matching can be done in simulation but in most cases iteration is required to get it right. Designing a filter that is not disturbed by the switching action of the ADC sample and hold circuitry is especially challenging.
Even something as routine as placing capacitors for supply decoupling affects overall performance and can cause board layout revisions. These tasks can easily consume months of engineering time debugging each revision and evaluating the changes.
Digital processing power offers several methods to improve the efficiency of base station PAs. Improvements in PA efficiency lead to significant reduction in the operating expense for cellular service providers.
Digital predistortion of the PA requires a low noise, high-speed receiver chain based on high performance analog components. In most cases, working with a high performance analog component adds complexity and requires more engineering time and resources. However, advances in integration such as the LTM9003 greatly simplify this task and allow the designer to focus on improving the digital algorithms.
Todd Nelson is manager of Signal Chain Module Development at Linear Technology Corp..