This paper analyzes the performance of MIMO broadcast channels with finite rate feedback. The key finding is that the feedback rate per mobile must be increased linearly with the SNR (in dB) to achieve the full multiplexing gain, which contrasts with point-to-point MIMO systems where no such increase is necessary. The analysis shows that for MIMO downlink channels, the feedback requirements are significantly higher than for point-to-point channels. The paper proposes a simple downlink transmission scheme using zero-forcing precoding with finite rate feedback. It demonstrates that the throughput of a feedback-based zero-forcing system is bounded if the SNR is taken to infinity and the number of feedback bits per mobile is kept fixed. The number of feedback bits per mobile must be increased linearly with the SNR to achieve the full multiplexing gain. The paper also shows that scaling the number of feedback bits according to B = α log₂ P for any α < M - 1 results in a strictly inferior multiplexing gain. The analysis concludes that the channel estimation error at the access point must scale as the inverse of the SNR to allow the full multiplexing gain to be achieved, resulting in the required linear scaling of feedback. The paper also discusses the use of random vector quantization (RVQ) for quantizing the channel, which is shown to perform measurably close to optimal quantization. The results show that the use of limited feedback leads to an SNR degradation relative to perfect CSIT. The paper also shows that the throughput of a feedback-based zero-forcing system can be bounded and that the multiplexing gain can be achieved by scaling the feedback rate linearly with the SNR. The analysis concludes that the throughput of a feedback-based zero-forcing system converges to the perfect CSIT throughput at asymptotically high SNR. The paper also discusses the performance of RVQ versus optimal vector quantization, showing that the RVQ penalty is actually very small. The paper concludes that the use of finite rate feedback in MIMO downlink channels is a practical approach that can achieve the full multiplexing gain when the feedback rate is appropriately scaled with the SNR.This paper analyzes the performance of MIMO broadcast channels with finite rate feedback. The key finding is that the feedback rate per mobile must be increased linearly with the SNR (in dB) to achieve the full multiplexing gain, which contrasts with point-to-point MIMO systems where no such increase is necessary. The analysis shows that for MIMO downlink channels, the feedback requirements are significantly higher than for point-to-point channels. The paper proposes a simple downlink transmission scheme using zero-forcing precoding with finite rate feedback. It demonstrates that the throughput of a feedback-based zero-forcing system is bounded if the SNR is taken to infinity and the number of feedback bits per mobile is kept fixed. The number of feedback bits per mobile must be increased linearly with the SNR to achieve the full multiplexing gain. The paper also shows that scaling the number of feedback bits according to B = α log₂ P for any α < M - 1 results in a strictly inferior multiplexing gain. The analysis concludes that the channel estimation error at the access point must scale as the inverse of the SNR to allow the full multiplexing gain to be achieved, resulting in the required linear scaling of feedback. The paper also discusses the use of random vector quantization (RVQ) for quantizing the channel, which is shown to perform measurably close to optimal quantization. The results show that the use of limited feedback leads to an SNR degradation relative to perfect CSIT. The paper also shows that the throughput of a feedback-based zero-forcing system can be bounded and that the multiplexing gain can be achieved by scaling the feedback rate linearly with the SNR. The analysis concludes that the throughput of a feedback-based zero-forcing system converges to the perfect CSIT throughput at asymptotically high SNR. The paper also discusses the performance of RVQ versus optimal vector quantization, showing that the RVQ penalty is actually very small. The paper concludes that the use of finite rate feedback in MIMO downlink channels is a practical approach that can achieve the full multiplexing gain when the feedback rate is appropriately scaled with the SNR.