Content for  TR 38.877  Word version:  18.1.0

For single-band FR2 radios, the application of DPD may not be critical compared to that of the FR1 counterpart. On one hand, the contribution of PAs to the total DC power consumption in a FR2 BS is much reduced. This is due to the fact that more antenna elements are added to increase the directivity of the antenna array to combat against the high path loss occurred at FR2, which in turn requires smaller power to feed each antenna element, and thus small-power PA per Tx is sufficient. On the other hand, the requirement on ACLR for FR2 BSs is not as large, i.e. 24 - 28dBc (by means of OTA measurement), compared to that applied to FR1 BSs, i.e. 45dBc [2] due to the beamforming and propagation environment. This somewhat alleviates the essential need of high-linearity PAs on meeting required ACLR. Therefore, the DPD in single-band FR2 BSs is expected to provide a little gain in terms of improving power efficiency and meeting the ACLR requirement. Furthermore, the analog and hybrid beamforming, which are predominantly used in FR2 radios, also pose challenges for DPD implementation. With hundreds to thousands of PAs and higher operating bandwidths for FR2 radio, simply utilizing a similar DPD architecture as in FR1 would cost extra for RF hardware design and power consumption, while it may be infeasible to deploy single DPD for every PA in the analog/hybrid beamforming phased array since several or all analog chains essentially share one digital path. The DPD algorithms would also have more demands on the bandwidth of feedback receiver/ADC and BB signal processing resources, which are scaled with the size of bandwidth to be linearized. These limited-gain and implementation-challenge factors make the similar DPD implementations as in FR1 less attractive to the FR2 single-band BS. Nevertheless, DPD may still bring benefit for FR2 BSs in terms boosting the overall system performance (e.g. throughput due to improved EVM, or very possibly energy efficiency). It is worth highlighting that common FR2 transmitter architecture requires to have tight integration between RF components to reduce hardware costs, sizes, and power loss in which isolator between an antenna element and a PA is preferably avoided [5], e.g. as illustrated in Figure 5.2.2.1-1. In such architecture, PAs directly interact with the antenna array due to low path isolation between them. As a result, mutual coupling and antenna mismatch between antenna paths have strong impact to the PAs' output matching impedance which changes the PA's efficiency and nonlinearity behaviours [5]. The array steering angle, which also alters the antenna matching impedance, shows strong dependence on the nonlinearity to the PAs too. In addition, input of PAs in different branches may be driven with different power as a result of the beamforming techniques applied (i.e. tapering, ZF, etc.), or gain error of the phase shifters and gain imbalance of power division network [6]. These mentioned factors have the detrimental effect to efficiency and linearity behaviours of PAs which may degrade the ACLR while increasing the OOB emission and beam distortions [5]. For example, several studies have demonstrated the impact of steering angle to ACLR and OOB emission and how DPD can help to improve the beamforming performance [5, 6, 7, 8, 9].

Obviously, any individual variation in PA would affect the performance of the linearization so the peak linearization performance is likely to be lower than that of a one-DPD-per-PA system. Whilst the design of PA's is likely to be identical there are a number of factors which could change their performance:

• Temperature - the location of each PA in the array (and the silicon) may mean different transistors are at different temperature (depending on the number of neighbouring devices for example) so the temperature of each junction may be different.
• Unit to unit variation - whilst some transistors may all be on a single piece of silicon and variation on a single bit of silicon may be small if multiple devices are used (8 or 16 per device may be more usual) so there will be unit to unit variation across the potential 128 paths
• Output match variation - PA performance is very dependent on the load it is working into, again all output match circuits are likely to be designed identically but will vary based on a number of factors:
  • Unit to unit of components
  • Antenna unit to unit
  • Antenna isolation and load pulling from nearby antenna (and signals)

Despite these issues useful linearization of an FR2 systems can be achieved and may become more common as technology improves.

Antenna array and TRXs in FR2 radios are desired to be tightly integrated in which RF filter is preferably omitted after the PA. Since a multi-band RIB essentially needs to transmit multiple-band signals concurrently, the nonlinearity of PAs will likely cause intermodulation (IM) distortion. It would be highly challenging to manage the PAs in multiband RIB not to operate in the nonlinearity power region. Particularly, the varying nonlinearity behaviours of the PAs, which causes by the nonlinear interaction between antenna array and the PAs as discussed above, also inherit to the FR2 multi-band RIB. Such issues would be expected to be more complicated in the multiband use cases than in single-band ones due to higher requirements on matching load impedance of PAs covering multiband/wideband. Therefore, unwanted emissions due to IM distortions likely exist and may possibly fall into the operating bands or inter-RF bandwidth. Note that the latter case occurs if there is multicarrier transmission in one band. Figure 5.2.2.2-1 and Figure 5.2.2.2-2 illustrate some examples. Assume that frequency ranges that the BS can operate in Band A and B are 24.25-26.5GHz (n258) and 27-29.5 GHz (n257), and there is transmission taking place at 26GHz in Band A and 27.5GHz in band B. Then IM3 components occur at 24.5GHz and 29GHz, which obviously fall into operating bandwidth of both bands as seen in Figure 5.2.2.2-1. Now assume that band A transmits two carrier frequencies at 24.75 and 25.75GHz. Then one IM3 component occurs at 26.75GHz which falls into the inter-RF bandwidth of the multiband BS as seen in Figure 5.5.2-3.

Note that Figure 5.2.2.2-1 and Figure 5.2.2.2-2 demonstrate the location of IMD products with CW signals, in reality the transmitted signals are wide band and the IMD products even more so, as such they may not be separable from the noise floor. As such IMD products are a potential issue but possibly not a serious one in some case, for example:

• 128 PA's at 20mW each is only 2.56W (34dBm)
• With 28dBc ACLR this gives an adjacent channel power of 6dBm

For a 100MHz channel this gives a PSD of -14dBm/MHz which is below the spurious emissions requirements for FR2. It is unlikely that any in-band non-linearities will be greater than the ACLR level (as these are 3rd order products) so the out of band emissions requirements are unlikely to be a problem. The IM distortion is also beamformed [10], however if the beams in the different bands are steered in different directions the IM product is in a different direction again [3]. If the beams are close to each other however, e.g. when UEs are nearby each other, the IM distortion beam may still be close enough to point at the intended UE. Thus, it needs to be managed to ensure that RF requirements are still be met for the FR2 multiband RIB. Apart from other UEs within the network, IM products in the inter RF bandwidth gap need to be suppressed sufficiently to ensure that spurious emissions requirements towards other systems are met. This would in particular need attention if band combinations between frequency groups would be considered. RF filter/diplexer after every PAs could be used to handle IM distortions. However, such solution would be very expensive since beamforming phased array in FR2 could have thousands of antenna paths. Filter per path may also generate significantly phase error between antenna paths and increase power loss. Thus, this may not be feasible in terms of cost, size, and performance of FR2 radios. Alternatively, DPD may be a cheaper solution for RF hardware architectures to this issue. For FR2 multi-band DPD, it essentially inherits the abovementioned advantages as well as challenges for single-band FR2 BS with a few additional issues. For instance, the very large percentage BW may cause potential issues when applying DPD for multi-band FR2, these are:

• The large percentage BW of the PA and the wider the BW the more difficult it will be to maintain a consistent impedance match of PAs over the entire band and hence memory effects may be greater making the DPD algorithm tougher to achieve good linearization.
• Larger BW means variation of the potential sources (as listed for single band) are greater, once gain reducing the potential linearity saving.

Potentially split or stacked element arrays mean each band signal may be fed to a different antenna, meaning the output load for the PA is more complex and more open to variation. Investigating practically feasible DPD solutions for beamforming phased array in FR2 has been an active research topic [11]. To address the concerns on the DPD implementation in FR2, DPD architectures and efficient DPD training model/algorithms have been intensively studies. Note that since one digital path will be shared between some or all analog chains in the analog/hybrid beamforming phased array, an architecture in which DPD linearization is applied for a set of PAs has been proposed and then demonstrated to be able to improve ACLR and EVM, e.g. [12, 13]. For multi-band DPD, as the beam steering for a multi-band FR2 system is likely to be applied to each band separately before the signals are combined and fed to a single PA, one example architecture could be that the signals will be generated separately in different converters. Such architectures have been investigated in FR1 bands [14] where separate signals were used to linearize a dual-band signals in a dual-band PA. In general, one can either deploy a common DPD for all bands or a dedicated DPD for each band [15]. The former may have less demand on DPD architecture but requires much higher bandwidth for DPD hardware as the captured, training and correcting samples are wideband; since the correction is wide-band, band-specific linearization may not be achievable as such. The latter has lower demand on DPD hardware bandwidth and can achieve per-band linearization but may require more complex DPD architectures and algorithms, i.e. to decompose the captured multiband signals to single-band one and vice versa for the corrections, as well as jointly optimize linearized coefficients for all bands. In both architectures, a wider bandwidth feedback path is required to deal with large signal bandwidth of FR2 band; traditionally the feedback channel needs 3 to 5 times the signal bandwidth to collect nonlinear information of PA. This can be a big burden for high-speed and high-precision ADCs. Nevertheless, DPD training algorithms which minimize the demand on DPD hardware bandwidth and BB resources have also been proposed, e.g. in [16, 17]. It should be highlighted that the implementation of phase shifters will not significantly impact the DPD solutions as linearization are applied to signals seen at the output of the PA, as mentioned in [18].