L
Admin
[Technical article, rfforum Editorial Team]
Almost every macro and small-cell base station PA shipped in the last decade is a Doherty. The reason is simple: modern OFDM signals have a peak-to-average power ratio of 7–10 dB, and a conventional class-AB PA is efficient only near its peak power. This note explains how the Doherty fixes that and where it gets difficult.
1. The architecture
Two amplifiers share one load. The main (carrier) amplifier is biased in class AB and runs all the time. The peaking amplifier is biased in class C and turns on only near the top of the power range. The main amplifier's output goes through a quarter-wave line with characteristic impedance Z0 = Ropt, which acts as an impedance inverter. The common load is Ropt/2 for a symmetric design.
2. Load modulation in numbers
At low drive the peaking amplifier is off. The main amplifier sees the inverter transforming Ropt/2 into 2·Ropt. With twice the optimum load it reaches voltage saturation at half its maximum current, so it is already at peak efficiency while delivering one quarter of the total peak power, which is 6 dB below peak. As the drive rises, the peaking amplifier injects current into the common node, the apparent load seen through the inverter falls, and at full power both amplifiers see Ropt.
Ideal class-B efficiency is π/4 ≈ 78.5 % at both points (peak and 6 dB back-off), and stays high in between. A single class-B amplifier at 6 dB back-off would be at only about 39 %.
3. Deeper back-off: asymmetric Doherty
For a signal with 8–10 dB PAPR, the 6 dB point is too shallow. Making the peaking device larger moves the second efficiency peak: the back-off level is 20·log10(1/γ), where γ is the main device's share of the peak current. With a peaking device twice the size of the main device (γ = 1/3) the peak moves to about 9.5 dB back-off. The cost is a deeper efficiency dip between the two peaks and more difficult gain flatness.
4. Why bandwidth is the hard part
The quarter-wave inverter is exactly a quarter wave at one frequency only. Away from the centre, load modulation becomes imperfect, and the offset lines that compensate the transistor's output parasitics are also frequency dependent. Classic Doherty designs therefore typically cover around 10–20 % fractional bandwidth. Wideband variants (parasitic-absorbing inverters, post-matching, sequential or load-modulated balanced amplifiers) are an active research area.
5. Practical issues
- Phase alignment: the two branches must combine in phase at the common node. A few tens of degrees of error costs both output power and efficiency, so the input splitter usually includes a tunable or offset line.
- Peaking turn-on: a class-C stage turns on softly, so the peaking current is lower than ideal in the transition region. Uneven input drive (giving more to the peaking branch) is a common fix.
- Linearity: the gain and AM-PM change as the peaking amplifier turns on. Doherty PAs almost always need digital predistortion (see https://www.rfforum.com/t/23).
6. A worked example
Target: 40 W (46 dBm) peak, 8 dB PAPR signal, average output about 38 dBm (6.3 W). A symmetric Doherty using two devices of about 20 W each puts the average power near (but slightly below) the 6 dB efficiency peak. An asymmetric 1:2 design would place the peak closer to the average power, at the cost of more tuning effort.
Discussion: for 3.5 GHz designs, have you found asymmetric Doherty worth the extra tuning compared with a symmetric design plus DPD? What drain efficiency at average power do you actually get once DPD is on?
— rfforum 编辑部 / Editorial Team
Almost every macro and small-cell base station PA shipped in the last decade is a Doherty. The reason is simple: modern OFDM signals have a peak-to-average power ratio of 7–10 dB, and a conventional class-AB PA is efficient only near its peak power. This note explains how the Doherty fixes that and where it gets difficult.
1. The architecture
Two amplifiers share one load. The main (carrier) amplifier is biased in class AB and runs all the time. The peaking amplifier is biased in class C and turns on only near the top of the power range. The main amplifier's output goes through a quarter-wave line with characteristic impedance Z0 = Ropt, which acts as an impedance inverter. The common load is Ropt/2 for a symmetric design.
2. Load modulation in numbers
At low drive the peaking amplifier is off. The main amplifier sees the inverter transforming Ropt/2 into 2·Ropt. With twice the optimum load it reaches voltage saturation at half its maximum current, so it is already at peak efficiency while delivering one quarter of the total peak power, which is 6 dB below peak. As the drive rises, the peaking amplifier injects current into the common node, the apparent load seen through the inverter falls, and at full power both amplifiers see Ropt.
Ideal class-B efficiency is π/4 ≈ 78.5 % at both points (peak and 6 dB back-off), and stays high in between. A single class-B amplifier at 6 dB back-off would be at only about 39 %.
3. Deeper back-off: asymmetric Doherty
For a signal with 8–10 dB PAPR, the 6 dB point is too shallow. Making the peaking device larger moves the second efficiency peak: the back-off level is 20·log10(1/γ), where γ is the main device's share of the peak current. With a peaking device twice the size of the main device (γ = 1/3) the peak moves to about 9.5 dB back-off. The cost is a deeper efficiency dip between the two peaks and more difficult gain flatness.
4. Why bandwidth is the hard part
The quarter-wave inverter is exactly a quarter wave at one frequency only. Away from the centre, load modulation becomes imperfect, and the offset lines that compensate the transistor's output parasitics are also frequency dependent. Classic Doherty designs therefore typically cover around 10–20 % fractional bandwidth. Wideband variants (parasitic-absorbing inverters, post-matching, sequential or load-modulated balanced amplifiers) are an active research area.
5. Practical issues
- Phase alignment: the two branches must combine in phase at the common node. A few tens of degrees of error costs both output power and efficiency, so the input splitter usually includes a tunable or offset line.
- Peaking turn-on: a class-C stage turns on softly, so the peaking current is lower than ideal in the transition region. Uneven input drive (giving more to the peaking branch) is a common fix.
- Linearity: the gain and AM-PM change as the peaking amplifier turns on. Doherty PAs almost always need digital predistortion (see https://www.rfforum.com/t/23).
6. A worked example
Target: 40 W (46 dBm) peak, 8 dB PAPR signal, average output about 38 dBm (6.3 W). A symmetric Doherty using two devices of about 20 W each puts the average power near (but slightly below) the 6 dB efficiency peak. An asymmetric 1:2 design would place the peak closer to the average power, at the cost of more tuning effort.
Discussion: for 3.5 GHz designs, have you found asymmetric Doherty worth the extra tuning compared with a symmetric design plus DPD? What drain efficiency at average power do you actually get once DPD is on?
— rfforum 编辑部 / Editorial Team
